<?xml version="1.0" encoding="UTF-8"?><rss xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:content="http://purl.org/rss/1.0/modules/content/" xmlns:atom="http://www.w3.org/2005/Atom" version="2.0" xmlns:cc="http://cyber.law.harvard.edu/rss/creativeCommonsRssModule.html">
    <channel>
        <title><![CDATA[Stories by Kavya Rathi on Medium]]></title>
        <description><![CDATA[Stories by Kavya Rathi on Medium]]></description>
        <link>https://medium.com/@kavyarathi1102?source=rss-e9b9ce957cf4------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/0*kiE-P8g8QrSjNJvp</url>
            <title>Stories by Kavya Rathi on Medium</title>
            <link>https://medium.com/@kavyarathi1102?source=rss-e9b9ce957cf4------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Sat, 21 Nov 2026 03:14:05 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@kavyarathi1102/feed" rel="self" type="application/rss+xml"/>
        <webMaster><![CDATA[yourfriends@medium.com]]></webMaster>
        <atom:link href="http://medium.superfeedr.com/" rel="hub"/>
        <item>
            <title><![CDATA[I Was Eight When I First Arrived Here. I’m Seventeen Now.]]></title>
            <link>https://medium.com/@kavyarathi1102/i-was-eight-when-i-first-arrived-here-im-seventeen-now-33f533ebc372?source=rss-e9b9ce957cf4------2</link>
            <guid isPermaLink="false">https://medium.com/p/33f533ebc372</guid>
            <category><![CDATA[memories]]></category>
            <category><![CDATA[high-school]]></category>
            <category><![CDATA[boarding-school]]></category>
            <category><![CDATA[schools]]></category>
            <dc:creator><![CDATA[Kavya Rathi]]></dc:creator>
            <pubDate>Wed, 18 Nov 2026 09:07:48 GMT</pubDate>
            <atom:updated>2026-11-18T09:07:48.528Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*H93cIjFpQFmT3JeMOisxsQ.jpeg" /><figcaption>Brighton College</figcaption></figure><p><em>Nine years, several versions of myself, friendships I never expected to find, and the rather difficult business of growing up somewhere you will eventually have to leave.</em></p><p>I cried on my first day.</p><p>Not particularly gracefully, either. I remember being eight years old, standing in a place that was supposed to become my home for the foreseeable future, wondering why on earth my parents thought this was a good idea. I had brought far too many things, understood very little of what was happening, and was entirely unconvinced by the assurances that I would be absolutely fine.</p><p>I was not absolutely fine.</p><p>I remember the rather overwhelming business of being introduced to people whose names I immediately forgot, being shown around places I could not yet distinguish from one another, and trying to understand how everyone else seemed to know exactly what they were doing. There were older students walking around as though they owned the place, people laughing in groups, and an entire world of routines that I had somehow been expected to understand within a few hours.</p><p>I remember wanting my parents to stay a little longer. Then wanting them not to leave at all. Then being embarrassed that I was crying, which, in retrospect, was a fairly ambitious expectation to have of an eight-year-old.</p><p>I don&#39;t remember the exact words anyone said to make me feel better. I do remember the feeling of that first evening: the unfamiliar room, the unfamiliar faces, and the very strange realisation that I was going to have to wake up in the same place the following morning.</p><p>I had no idea what I was beginning.</p><p>If somebody had told me that I would still be here at seventeen, that I would eventually find myself watching other people leave with the same tears I had once tried so desperately to hide, I would have found the suggestion quite ridiculous.</p><p>Nine years is a difficult amount of time to comprehend when you are eight. It is almost your entire life. At that age, eighteen seems impossibly far away, and the older students appear to belong to an entirely different category of human being. They are taller, more confident, seem to know everyone, and have an almost irritating ability to make everything look effortless.</p><p>I remember looking at the graduating students and thinking they were practically adults. They seemed so certain of themselves. They had their friendships, their routines, their places in the school, and an understanding of the world I assumed I would acquire automatically when I got older.</p><p>I never considered that they might have been just as uncertain as I was.</p><p>Now I am seventeen, and I am watching the next group of younger students look at the older ones in precisely the same way.</p><p>It is a rather strange position to find yourself in.</p><p>I have spent so much of my childhood here that I sometimes forget how much of it has happened within the same few buildings, among many of the same people. The place has witnessed me at eight, nine, ten, eleven, twelve, thirteen, fourteen, fifteen, sixteen and now seventeen. It has known versions of me that I barely recognise anymore.</p><p>And the people who have been here alongside me have witnessed all of them.</p><p>I think that is what makes the experience difficult to explain to someone who has never boarded. It is not simply that you go to school together. You grow up together, often without realising how much of your growing up is being witnessed by everyone around you.</p><p>You remember each other&#39;s awkward phases, terrible decisions, questionable haircuts, changing interests, first achievements, disappointments and the many small embarrassments that would otherwise have disappeared into childhood. There are people here who remember things about me from when I was eight that I have completely forgotten myself. Unfortunately, they are also quite happy to remind me of them.</p><p>There is something both lovely and deeply inconvenient about having friends who have known you for that long.</p><p><strong>The people who made the place feel like home</strong></p><p>I think the first thing I learnt about boarding school was that you could miss home terribly and still find another kind of home somewhere else.</p><p>It took time, of course. I don&#39;t think you arrive somewhere on a Monday and decide by Friday that you belong there. At least, I didn&#39;t. In the beginning, everything was unfamiliar. Even the smallest things required an adjustment. You had to learn where things were, who to ask, how the days worked and, more importantly, how to find your place among people who had already found theirs.</p><p>Then you meet someone.</p><p>Sometimes it is the person who happens to sit next to you at a meal. Sometimes it is someone who offers to show you around because they remember what it was like to be new. Sometimes it is a person you have absolutely nothing in common with on paper, but who, for reasons neither of you could properly explain, becomes part of your everyday life.</p><p>I have always found it quite amusing that some of the people who have become most important to me were not people I would necessarily have imagined becoming friends with when I first met them.</p><p>You start by asking someone a question. Then you begin sitting together. Then you find yourself looking for them whenever something happens. Eventually, you stop making plans to see each other because you already know you will.</p><p>And somewhere along the way, they become family.</p><p>Not in the dramatic sense of the word, and certainly not because everyone suddenly gets along. Quite the opposite, actually. You argue. You irritate one another. You have disagreements about things that seem extraordinarily important at the time and completely insignificant a week later. You learn each other&#39;s habits, including the ones you would very happily change. You discover that some people are absolutely incapable of being ready on time, that others will borrow your things and forget to return them, and that there are friendships in which the phrase &quot;I am not speaking to you&quot; is rarely a particularly convincing threat.</p><p>But you also learn what it means to have people around you without having to ask.</p><p>There is a particular comfort in knowing that someone will notice if you are unusually quiet at dinner. That someone will knock on your door simply because they have nothing better to do. That there will be a person who knows when you need to talk and another who understands that you would much rather be distracted.</p><p>You learn to recognise the sound of a friend&#39;s knock. You know whose room you can walk into without an invitation and whose door you should probably not open without checking first. You know who will be awake when you cannot sleep, who will make you laugh when you are in a terrible mood, who will lend you something without asking why you need it, and who will tell you, quite bluntly, when you are being unreasonable.</p><p>You also learn that friendship is not always particularly photogenic.</p><p>Sometimes it is sitting on somebody&#39;s floor while they finish getting ready. Sometimes it is accompanying someone to do something incredibly mundane because neither of you wants to be alone. Sometimes it is sharing food, lending clothes, helping someone find something they have misplaced for the third time that day, or simply staying in the same room while doing entirely separate things.</p><p>There are friendships that are built through extraordinary experiences, but I think some of the strongest ones are built through repetition. Seeing someone every day, in every possible mood, until their presence becomes one of the most ordinary and reassuring parts of your life.</p><p>I have friends here who know my family, my habits, my history and probably more about my childhood than I would like them to. They have seen me grow up, and I have had the privilege of seeing them do the same.</p><p>We have celebrated one another&#39;s successes, been there for the less successful occasions, and accumulated an almost unreasonable number of memories that would make very little sense to anyone outside our group.</p><p>There are jokes that have been running for years. Stories that have been retold so many times that the original version has probably been lost. Nicknames that have somehow survived despite repeated objections. Conversations that begin with absolutely nothing and end with somebody saying something that will be quoted for the next six months.</p><p>And then there are the quieter things.</p><p>The person who notices you have not eaten properly. The friend who sits beside you when you are having a difficult day without insisting that you explain it. The one who remembers something you mentioned weeks ago. The one who makes sure you are included when everyone else seems to have forgotten.</p><p>I don&#39;t think people always realise how much these small gestures matter.</p><p>When you live with your friends, you see their lives beyond the version they choose to show everyone else. You see them when they are tired, homesick, stressed, disappointed, excited, frightened or simply having an ordinary day. You become familiar with the parts of them that do not necessarily make it into conversations or photographs.</p><p>And they see the same parts of you.</p><p>It is not always comfortable. In fact, it can be rather humbling. You cannot maintain an elaborate version of yourself when the people around you have watched you spend an entire morning looking for something that was in your hand all along.</p><p>But there is something reassuring about being known in that way.</p><p>I think that is one of the things I will miss most. Not just having friends, but having people who know the context behind almost everything. People who do not need the entire explanation because they were there when it happened. People who can look at you across a room and understand exactly what you are thinking, which is occasionally useful and occasionally extremely inconvenient.</p><p>I have come to realise that family does not always have to be defined by where someone comes from. Sometimes it is defined by the number of ordinary days you have spent together, the things you have been through, and the fact that you have continued to choose one another.</p><p>Somewhere between eight and seventeen, a great many people stopped being people I went to school with and became people I cannot quite imagine my life without.</p><p>I am not sure when that happened.</p><p>I am very glad it did.</p><p><strong>The people who leave before you do</strong></p><p>One of the more peculiar parts of growing up in a boarding school is watching people leave.</p><p>When you are younger, the graduating students seem like an entirely different species. You know they will eventually finish school, of course, but the idea is so distant that it barely feels relevant. They are simply the older ones. They are the people who know everything, who seem to have the most interesting lives, and who occupy a kind of position within the school that you cannot quite imagine yourself reaching.</p><p>Then one year, they are gone.</p><p>I remember watching my first group of seniors graduate. I remember the photographs, the celebrations, the speeches and the general excitement surrounding the occasion. I remember thinking how wonderful it must be to be at that stage of life, to have finished something so significant and to be moving on to whatever came next.</p><p>I also remember being rather confused by how upset everyone seemed.</p><p>I understood that people were sad. I just didn&#39;t understand the full extent of it. Surely everyone would keep in touch? Surely they would come back? Surely leaving school did not mean that you were suddenly no longer part of it?</p><p>I was too young to understand that a place can remain exactly where it is while your relationship with it changes completely.</p><p>I understand it rather better now.</p><p>By the time you have spent several years here, the people leaving are no longer simply older students you recognise from around campus. They are friends. They are people you have shared meals with, people whose rooms you have spent hours in, people who have been part of your routines for years. Some of them have been around for almost as long as you can remember. Others have entered your life more recently but somehow managed to become equally important.</p><p>You watch them prepare for their final examinations, talk about university, make plans and gradually begin to occupy themselves with the world outside school. You are happy for them. You want them to succeed. You are excited to see what they do next.</p><p>And then graduation comes around.</p><p>I don&#39;t think I was prepared for the first time I watched someone I genuinely loved leave.</p><p>There is something particularly difficult about the last few weeks of a graduating year. Everyone knows what is coming, but nobody quite knows how to behave about it. People become sentimental over things they would normally complain about. A familiar routine suddenly seems worth preserving. Plans are made with the assumption that there will be another opportunity, even when everyone knows that the circumstances will never be quite the same again.</p><p>There are photographs everywhere. People are writing messages, exchanging gifts, making promises and trying to fit an unreasonable number of memories into a few remaining days.</p><p>And, eventually, someone cries.</p><p>Then somebody else does.</p><p>Then the people who had been insisting all week that they were perfectly fine discover that they are not.</p><p>I have watched people I considered almost permanent parts of the school pack their belongings and leave. I have stood with friends while they said goodbye to people who had become their family. I have watched younger students look at graduating seniors with the same admiration I once had, and I have seen people who were usually very composed struggle to get through a goodbye.</p><p>It is difficult not to be affected by it.</p><p>The worst part, I think, is that you do not always know which goodbye will stay with you. Sometimes it is the person you expected to be upset about leaving. Sometimes it is someone whose departure you had not properly considered until you realised you would no longer see them every morning.</p><p>You tell them you will visit. They tell you they will come back. Everyone promises to keep in touch. There are plans for reunions, holidays, dinners and all the things people genuinely intend to do.</p><p>And perhaps they will.</p><p>But you also know that life is about to become considerably less convenient. People will move to different countries, attend different universities, make new friends and begin lives that no longer revolve around the same timetable. The person who used to be three doors away will suddenly be a flight away.</p><p>It is not that the friendship has to end. It is that the circumstances that made it so effortless are ending.</p><p>That distinction took me a while to understand.</p><p>I have also learnt that watching people leave changes the way you look at your own time here. You begin to notice the younger students who are still at the beginning of their school lives. You realise that the people you currently see every day will eventually be the ones someone else is saying goodbye to.</p><p>And, rather unexpectedly, you find yourself becoming one of the people you once cried over losing.</p><p><strong>Somewhere along the way, I became one of the older students.</strong></p><p>I don&#39;t think I noticed it happening.</p><p>There was no formal announcement that I had crossed some invisible threshold. One year I was the younger student trying to work out where I belonged, and then, gradually, I became someone who was expected to know the answers.</p><p>You start recognising the new students. You notice when someone is sitting alone. You know which parts of the school are confusing when you first arrive and which traditions take a while to understand. You become aware of the fact that somebody younger might be watching you in the same way you once watched the older students.</p><p>I find that quite a responsibility.</p><p>I remember how much the smallest kindnesses mattered when I was eight. Someone remembering my name, making room for me, explaining something without making me feel silly for not knowing it, or simply being friendly when I was clearly having a difficult day.</p><p>I do not remember every person I met during my first few weeks. But I remember how it felt when someone made the place seem a little less unfamiliar.</p><p>I hope I have managed to do that for someone else.</p><p>There is also a rather amusing side to becoming one of the older students. You begin to understand the things that probably annoyed the people who came before you. You find yourself saying things you distinctly remember hearing when you were younger. You become more conscious of time, responsibilities and the fact that the younger ones have absolutely no intention of listening to you until they are ready.</p><p>And then you hear yourself giving advice and realise that you have become precisely the kind of person you once thought was impossibly grown up.</p><p>I am still not convinced I am particularly qualified.</p><p>But I suppose that is how it works. You spend years being looked after by people who are slightly older than you, and eventually you find yourself looking after people who are slightly younger. The responsibility passes quietly, without anyone asking whether you feel ready for it.</p><p>I think there is something rather special about that continuity. The students change, the years pass, and the people who once made the school feel like home eventually leave it. But they leave behind small things: traditions, habits, stories, advice, and ways of treating people that are carried forward by those who remain.</p><p>You realise that you are part of something that existed before you arrived and will continue long after you leave.</p><p>For a long time, I thought of the school as something that belonged to the people who had been there the longest. I now understand that everybody gets a turn at belonging to it, even if only for a few years.</p><p>And I think that is rather beautiful.</p><p><strong>There are nine years of memories I cannot possibly fit into an article.</strong></p><p>I could write about the celebrations, the school events, the trips, the competitions, the traditions and the occasions that everyone photographs. I could write about the friendships that began in the most unlikely circumstances, the people who became inseparable, and the memories that still make me laugh when I think about them.</p><p>I could write about the ordinary evenings that turned into something else entirely because someone decided to come over and nobody quite wanted to go to bed. About the meals that lasted longer than they should have, the last-minute preparations, the conversations that continued long after they had run out of any sensible subject, and the collective confusion that seems to accompany almost every school event.</p><p>I could write about the way a group of people from entirely different backgrounds can develop a language of their own. The references nobody else understands, the little habits that become traditions, and the strange familiarity of knowing exactly what somebody is going to say before they say it.</p><p>I could write about how quickly you become accustomed to having people around. How you can be surrounded by hundreds of students and still have a handful of people who make the entire place feel smaller, safer and more familiar.</p><p>But I think the memories that matter most are not necessarily the ones that would make for an interesting article.</p><p>They are the moments that nobody thought to photograph.</p><p>The conversations that happened because someone happened to be awake at the same time as you. The occasions when a friend stayed with you without needing to know exactly what was wrong. The laughter that came at the most inappropriate possible moment. The days when nothing particularly exciting happened, but you were happy because the people you wanted around were there.</p><p>And the people.</p><p>Always the people.</p><p>I have friends here who have seen me change in ways I probably could not have managed to explain to them at the time. Some have known me since I was little. Others have arrived at different stages of my life and somehow become just as important. There are friendships that have survived changes in year groups, different routines, disagreements, distance during holidays and all the other things that make growing up slightly complicated.</p><p>There are people who know my family and people whose families I have come to know. People whose homes have become familiar to me, and whose parents have somehow become people I am comfortable speaking to. There are friends whose younger siblings I have watched grow up, just as they have watched me grow up.</p><p>It is a strange thing to realise that your childhood has become intertwined with so many other people&#39;s childhoods.</p><p>I sometimes think about what it will be like to meet these people again in ten or fifteen years. What we will have become, where we will live, what we will be doing, and whether we will still laugh at the same ridiculous things. I wonder whether we will remember the details of certain days or whether our memories will become a little less precise with time.</p><p>I hope we remember enough.</p><p>I hope that, when we meet again, we will still be able to sit together and speak as though no time has passed. I hope we will remember the people who were there, the things we did and the versions of ourselves that existed within these walls.</p><p>And I hope we will be able to laugh at how seriously we took certain things.</p><p>Because we did take certain things very seriously.</p><p><strong>I am seventeen, and for the first time, I can see the end.</strong></p><p>There is something rather uncomfortable about knowing that you are approaching the last chapter of something while you are still living inside it.</p><p>I am not leaving tomorrow. There is still time. There are examinations to sit, things to organise, people to spend time with and a great many ordinary days left to enjoy. I have no intention of spending the entirety of my final year behaving as though every occasion is a farewell.</p><p>That would be exhausting, and probably rather unpleasant for everyone around me.</p><p>But I would be lying if I said I had not started thinking about it.</p><p>I notice things now that I probably would not have noticed a few years ago. I pay more attention to the people around me. I find myself appreciating routines I once complained about. I am more aware of the fact that the people I see every day will eventually become people I have to make plans to see.</p><p>I have started to understand why the older students used to look so emotional during their final weeks.</p><p>I understand why people take so many photographs, why they write messages in one another&#39;s books, why they insist on making plans to meet again, and why even the most ordinary goodbye can become unexpectedly difficult.</p><p>I understand why someone can spend an entire afternoon insisting that they are excited to leave, only to find themselves crying when the actual moment arrives.</p><p>I understand because I have been on the other side of it.</p><p>I have watched seniors leave while crying myself. I have seen friends try to be brave for one another, only to become emotional the moment they realise that the goodbye is actually happening. I have watched people who had spent years together suddenly have to consider what their friendship would look like without a shared campus, a shared timetable or the certainty of seeing one another the following morning.</p><p>And now I am beginning to realise that, before very long, somebody will be watching me leave.</p><p>I find that thought quite difficult.</p><p>Not because I am frightened of what comes next. I am genuinely looking forward to the next part of my life. There is so much I want to do, so much I want to learn and so many places I want to see. I am excited about university and the independence that comes with it. I want to discover who I am outside the environment I have known for so long.</p><p>But I also know that I will never again be eight years old, arriving at this school for the first time.</p><p>I will never have the experience of meeting these people for the first time, of slowly discovering who would become important to me, or of watching the older students and wondering what it would feel like to be their age.</p><p>I will never get those first days back.</p><p>And, perhaps more importantly, I will never get these particular days back either.</p><p>The ones I am living now.</p><p>It is quite easy to spend your school years waiting for the next thing. When you are younger, you want to be older. You want to be trusted with more responsibility, have more freedom, be included in the conversations of the older students and feel as though you have finally grown into the place.</p><p>Then you get there, and you begin looking back.</p><p>I wish I had understood that a little earlier. Not because I regret anything, but because there are days I would quite happily relive simply because I did not know they would become memories.</p><p>I would like to be eight again for an afternoon. I would like to see the place through the eyes of the child who arrived crying, who thought the older students were impossibly grown up and who had no idea that the people she had not yet met would eventually become so important.</p><p>I would like to tell her that she is going to be all right.</p><p>That she will find her people. That some of the people she meets will become the sort of friends she will still be speaking to years later. That she will have days when she misses home and days when she cannot wait to get back to school. That she will laugh until she cannot explain what was funny, have arguments that seem terribly important, make mistakes, learn to apologise and discover that growing up is rarely as straightforward as it looks from the outside.</p><p>I would tell her that one day she will be one of the older students. That she will watch the people she admires graduate, and that she will cry when they leave, even though she once thought she could never understand why anyone would be so upset.</p><p>I would tell her that the place she is so uncertain about will become familiar enough for her to miss it before she has even left.</p><p>And I would tell her to be a little less embarrassed about crying on that first day.</p><p>She had no idea what she was being given.</p><p><strong>Perhaps the hardest part of growing up somewhere is knowing that you cannot take everyone with you.</strong></p><p>I don&#39;t mean that friendships have to end when school does. Some of the people I love most will remain in my life, and I have every intention of making sure that they do. We will visit one another, plan holidays, arrange dinners, attend important occasions and, hopefully, continue to be involved in the unremarkable details of one another&#39;s lives.</p><p>But I know it will be different.</p><p>We will no longer have the luxury of being in the same place by default. We will have to make an effort. We will have different schedules, different responsibilities and different lives. There will be conversations we miss because of time differences, occasions we cannot attend and stretches of time when everyone is simply too occupied to speak as often as they would like.</p><p>That is part of growing up, I suppose.</p><p>I don&#39;t think it makes the friendships any less meaningful. If anything, it makes me appreciate the circumstances in which they were formed. There is something extraordinary about having spent so many years with people simply because life happened to put you in the same place.</p><p>I think of all the people who have been part of my life here, and I realise that I have been given something I could never have planned for myself. A collection of people, each with their own personalities, backgrounds, ambitions and stories, who somehow became part of my own.</p><p>Some of them have taught me things without ever knowing they were teaching me. Some have made difficult years easier. Some have challenged me, annoyed me, encouraged me and made me laugh when I needed it most. Some have become the people I call first when something happens, whether it is wonderful, terrible or so insignificant that it probably does not warrant a phone call.</p><p>They have become part of the way I understand friendship.</p><p>And I hope I have been able to give them even a fraction of what they have given me.</p><p>I don&#39;t know what the next few years will look like. I don&#39;t know which friendships will change, which people I will see every week and which I will only see occasionally. I don&#39;t know what it will feel like to wake up somewhere else and realise that the familiar faces around me are no longer the ones I have known for most of my childhood.</p><p>I suppose that is the point. For the first time in nine years, I am approaching something without knowing exactly what it will look like.</p><p>Perhaps that is what growing up is supposed to feel like.</p><p>I started this piece thinking about my first day. I thought I would remember a particular detail, find the right words for it and somehow explain what nine years had meant to me.</p><p>I am not sure I have managed that.</p><p>I still remember the eight-year-old who cried because she did not want her parents to leave. I remember how enormous everything seemed, how unfamiliar the place felt and how certain I was that I would never quite get used to it.</p><p>I wish I could tell her that she would.</p><p>I wish I could tell her about the friends she would make, the people who would become family, the older students she would admire and eventually become, and the younger ones she would one day find herself looking after. I wish I could tell her about all the people who would walk into her life without either of them knowing how important they would eventually become.</p><p>I would tell her that she would grow up here. Properly grow up. That she would change in ways she could not anticipate, and that the people around her would change too. That there would be people she would know for almost her entire childhood, and others who would come into her life much later but somehow feel as though they had always belonged there.</p><p>I would tell her that one day she would stand among the people she loved, watching them leave, and understand exactly why the older students had cried.</p><p>And I would tell her that, eventually, it would be her turn.</p><p>I am seventeen now. Eighteen is approaching, and so is the final stretch of a chapter that has occupied almost my entire life.</p><p>I don&#39;t think I am ready to say goodbye. I am not sure I am supposed to be.</p><p>There is a photograph somewhere of me on that first day. I am sure I look rather small, probably rather unhappy, and entirely unaware of the life that was about to unfold around me. I would like to put it beside a photograph of myself now. Nine years between the two. Same person, supposedly, although I suspect the eight-year-old would have quite a lot to say about the seventeen-year-old.</p><p>I wonder what she would think of the people. I wonder which of my friends she would have liked immediately, which ones she would have found intimidating, and which ones she would never have imagined becoming so close to. I wonder what she would make of the fact that the older students she once thought were practically adults are now people she calls her friends.</p><p>Most of all, I wonder whether she would recognise this place.</p><p>I think she would. Not because everything looks exactly the same, but because she would recognise the feeling of it. The familiarity, the noise, the people, the peculiar sense that life is happening in several different rooms at once.</p><p>And perhaps she would be surprised to discover that the place she was so desperate to leave that first evening would become somewhere she would one day struggle to say goodbye to.</p><p>I arrived here at eight, crying because I did not want to stay.</p><p>At seventeen, I am trying not to cry because I know that, before long, I will have to leave.</p><p>Nine years ago, I had no idea that I was arriving somewhere that would become part of me.</p><p>Now I cannot quite imagine leaving without taking a little of it with me.</p><p>And I suppose that is what I have been trying to say all along.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=33f533ebc372" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[What happens if large language models reach a scaling plateau?]]></title>
            <link>https://medium.com/@kavyarathi1102/what-happens-if-large-language-models-reach-a-scaling-plateau-f1b4ae2a464c?source=rss-e9b9ce957cf4------2</link>
            <guid isPermaLink="false">https://medium.com/p/f1b4ae2a464c</guid>
            <category><![CDATA[ai-agent]]></category>
            <category><![CDATA[machine-learning]]></category>
            <category><![CDATA[llm]]></category>
            <category><![CDATA[ai]]></category>
            <dc:creator><![CDATA[Kavya Rathi]]></dc:creator>
            <pubDate>Tue, 30 Jun 2026 17:30:23 GMT</pubDate>
            <atom:updated>2026-06-30T17:30:23.265Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*17fOlPiYoVPBaPTd" /><figcaption>Photo by <a href="https://unsplash.com/@lukejonesdesign?utm_source=medium&amp;utm_medium=referral">Luke Jones</a> on <a href="https://unsplash.com/?utm_source=medium&amp;utm_medium=referral">Unsplash</a></figcaption></figure><h3>What Happens If AI Stops Scaling?</h3><p>For the better part of the last decade, the trajectory of artificial intelligence has appeared almost deceptively simple. Train larger models, feed them more data, provide more computational power, and performance improves. This relationship, formalised through scaling laws, transformed what had once been an unpredictable field into something that resembled an engineering discipline. Instead of searching endlessly for new algorithms, researchers increasingly discovered that larger neural networks, trained on larger datasets using greater computational resources, consistently exhibited capabilities that smaller models simply could not.</p><p>The results were extraordinary.</p><p>Language models progressed from producing fragmented sentences to writing coherent essays, solving programming problems, translating languages, generating software, creating realistic imagery, and assisting with scientific research. Capabilities that once appeared decades away seemed to emerge almost naturally from scale itself. Every successive generation of models appeared not merely incrementally better but qualitatively different, exhibiting behaviours that many researchers had not explicitly programmed into them.</p><p>For a time, the prevailing assumption became remarkably straightforward: intelligence was largely a function of scale.</p><p>Yet engineering history has repeatedly demonstrated that exponential progress rarely continues indefinitely. Every technology eventually encounters physical, economic, or architectural constraints that force its trajectory to change. Aircraft engines reached limits that demanded entirely new propulsion concepts. Semiconductor manufacturing continually confronts atomic-scale boundaries. Even Moore’s Law, once considered almost inevitable, has gradually slowed as transistor dimensions have approached fundamental physical limits.</p><p>Artificial intelligence may now be approaching a similar inflexion point.</p><p>Over the past several years, each successive frontier model has required dramatically larger investments in data, specialised hardware, electrical power, and financial capital. Training runs now involve tens of thousands of advanced GPUs operating continuously for weeks or months, consuming enormous quantities of energy while costing hundreds of millions of dollars. At the same time, the performance improvements delivered by these increasingly expensive training runs appear, at least in some domains, to be becoming less dramatic than previous generations.</p><p>This observation has led to an increasingly common question throughout both industry and academia.</p><p>Have large language models begun approaching a scaling wall?</p><p>The phrase itself has become something of a catch-all description for the possibility that simply increasing model size may no longer produce the transformative gains that characterised the previous decade of deep learning. Rather than representing the end of progress, it suggests that the relationship between computational investment and practical capability may no longer remain as favourable as it once was.</p><p>Whether that conclusion is correct remains far from settled.</p><p>There are compelling reasons to believe that today’s architectures still possess considerable untapped potential. Reinforcement learning continues to demonstrate impressive improvements in reasoning, planning, and tool use. Novel training objectives may allow models to extract substantially more value from existing datasets. Synthetic data generation offers one possible response to the growing scarcity of high-quality human-written text, while entirely new model architectures may eventually replace today’s transformer-based systems altogether.</p><p>History offers numerous examples where apparent technological ceilings proved temporary.</p><p>Aviation repeatedly surpassed perceived aerodynamic limits. Semiconductor engineers found innovative methods for extending Moore’s Law long after many experts predicted its demise. Scientific progress rarely proceeds along a perfectly smooth trajectory; periods of apparent stagnation often precede entirely unexpected breakthroughs.</p><p>For that reason, declaring the scaling era over would almost certainly be premature.</p><p>Nevertheless, hypothetical questions often prove surprisingly useful.</p><p>Suppose, for a moment, that we have indeed extracted most of the easily accessible gains from scaling contemporary language models. Imagine that larger datasets, more computational resources, and increasingly expensive training runs continue producing improvements- but only marginal ones. What would become of artificial intelligence if simply building ever-larger models ceased to be the dominant strategy?</p><p>Interestingly, such an outcome would not necessarily represent a crisis.</p><p>Instead, it might signal the beginning of a new phase in AI’s evolution.</p><p>Throughout the history of technology, periods of explosive discovery are frequently followed by periods of systems engineering. Once the foundational breakthrough has been achieved, progress shifts away from making the core invention fundamentally different and towards learning how to integrate, optimise, and deploy it more effectively.</p><p>The invention of the transistor did not immediately produce smartphones or cloud computing. Instead, decades of engineering transformed a relatively simple electronic switch into increasingly sophisticated integrated circuits. Likewise, the invention of packet switching did not instantly create the modern internet. The revolutionary applications that define today’s digital world emerged only after engineers spent decades building protocols, infrastructure, and software upon that original concept.</p><p>Artificial intelligence may now be approaching a comparable transition.</p><p>Rather than asking how to make one enormous model infinitely more intelligent, researchers may increasingly focus on how to organise existing intelligence into more capable systems.</p><p>Viewed from this perspective, the scaling wall becomes less of an obstacle and more of a change in direction.</p><p>Indeed, there is reason to believe that many of the limitations attributed to modern language models arise not because they lack intelligence altogether, but because they embody only one particular form of intelligence exceptionally well.</p><p>Large language models excel at recognising patterns across vast quantities of information. They synthesise knowledge, generate language with remarkable fluency, identify relationships, and produce responses that often appear deeply intuitive. Yet they also display familiar weaknesses: inconsistent logical reasoning, susceptibility to hallucinations, difficulty maintaining long chains of precise deduction, and occasional failures on problems that seem trivial to humans.</p><p>These characteristics suggest an intriguing possibility.</p><p>Perhaps the intelligence we have successfully built is not incomplete because it is too small, but because it resembles only one component of cognition rather than the entirety of it.</p><p>If that interpretation proves correct, then simply making today’s models larger may eventually yield diminishing returns- not because intelligence itself has stopped improving, but because scaling amplifies the same underlying strengths and weaknesses simultaneously.</p><p>The next frontier would therefore involve complementing these capabilities rather than merely enlarging them.</p><p>Understanding what that complementary intelligence might look like requires stepping away from computer science for a moment and examining the only genuinely intelligent system we currently know: the human mind.</p><p>Only by understanding how human cognition combines intuition with deliberate reasoning can we begin to appreciate why scaling alone may never have been the complete answer- and why the future of artificial intelligence may ultimately depend less on building larger brains than on building better cognitive systems.</p><h3>Two Kinds of Intelligence: Intuition and Deliberation</h3><p>To understand where artificial intelligence may be headed, it helps to step away from computers entirely.</p><p>For decades, AI research has largely measured progress by comparing machines to people. We ask whether a model can reason like a human, write like a human, or solve problems as effectively as a human expert. Yet despite the remarkable capabilities of today’s large language models, they still exhibit an odd contradiction. They can draft compelling essays, explain quantum mechanics, generate functioning software, and compose convincing poetry- only to occasionally fail at surprisingly simple logical tasks or basic arithmetic.</p><p>At first glance, this inconsistency appears almost paradoxical.</p><p>How can a machine capable of discussing advanced theoretical physics confidently misinterpret a straightforward numerical comparison? Why can an AI produce elegant legal arguments while occasionally making elementary factual errors? More importantly, why do these mistakes feel so fundamentally different from ordinary software bugs?</p><p>The answer may lie not in computer science, but in cognitive psychology.</p><p>One of the most influential frameworks for understanding human thought was developed by psychologist <strong>Daniel Kahneman</strong>, who proposed that our minds operate through two distinct yet deeply interconnected modes of thinking. Although these are not literally separate systems inside the brain, they provide an extraordinarily useful model for understanding how people make decisions.</p><p>The first mode operates almost effortlessly.</p><p>It recognises familiar faces in crowded rooms before we consciously begin looking. It completes familiar phrases without deliberate thought. It instantly understands sarcasm, detects emotion in someone’s voice, predicts where a moving object will land, and allows experienced drivers to navigate familiar roads while carrying on conversations.</p><p>This mode is rapid, intuitive, and automatic.</p><p>It relies on accumulated experience rather than explicit reasoning, constantly making predictions about the world based upon patterns it has encountered before. Most importantly, it functions without requiring conscious effort.</p><p>Psychologists commonly refer to this as <strong>System 1</strong>.</p><p>The second mode behaves very differently.</p><p>When solving a difficult mathematical proof, analysing a legal contract, learning a new language, or carefully evaluating competing scientific hypotheses, intuitive guesses are rarely sufficient. Instead, we consciously slow down, examine evidence, test assumptions, and work methodically through each step of the problem.</p><p>This process is deliberate rather than automatic.</p><p>It consumes mental energy, requires sustained attention, and often feels mentally exhausting after prolonged use. Yet it also enables careful logical reasoning, long-term planning, and the correction of intuitive mistakes.</p><p>This is known as <strong>System 2</strong>.</p><p>Human intelligence emerges not from either system individually, but from the interaction between both.</p><p>Most of everyday life is handled by intuition because intuition is remarkably efficient. If every decision required painstaking logical analysis, ordinary activities would become almost impossible. We do not consciously calculate the trajectory of a thrown ball before catching it, nor do we evaluate grammatical rules every time we speak.</p><p>Reasoning intervenes only when intuition becomes insufficient.</p><p>In effect, System 1 proposes answers, while System 2 evaluates whether those answers deserve to be trusted.</p><p>That relationship may offer one of the clearest ways to understand modern artificial intelligence.</p><h3>The First Era of AI: Building Pure Reason</h3><p>Curiously, early artificial intelligence research pursued almost the exact opposite strategy.</p><p>During the 1950s through the 1980s, researchers largely believed that intelligence arose through explicit logical reasoning. If human expertise could be translated into rules and those rules encoded into computers, machines would eventually become intelligent.</p><p>This philosophy became known as <strong>Good Old-Fashioned Artificial Intelligence (GOFAI)</strong>, or more broadly, symbolic AI.</p><p>Rather than learning from data, symbolic systems manipulated carefully defined symbols according to explicit logical rules. Engineers painstakingly encoded human knowledge into enormous collections of if-then statements designed to mimic expert reasoning.</p><p>Within well-defined environments, the results were genuinely impressive.</p><p>Chess programs analysed millions of legal positions using formal search algorithms. Mathematical theorem provers demonstrated elegant proofs. Expert systems such as <strong>MYCIN</strong> successfully assisted medical diagnosis by applying hundreds of carefully crafted clinical rules, while <strong>DENDRAL</strong> helped chemists identify molecular structures through symbolic reasoning.</p><p>These systems did not merely automate calculations.</p><p>They represented humanity’s first serious attempt to formalise intelligence itself.</p><p>Yet despite decades of research, symbolic AI repeatedly encountered the same fundamental limitation.</p><p>The real world refused to follow explicit rules.</p><p>Human language is ambiguous. Social interactions constantly violate simple logical structures. Everyday environments contain exceptions, uncertainty, incomplete information, and countless situations too numerous to anticipate individually.</p><p>Writing rules for every possible circumstance proved essentially impossible.</p><p>As the number of exceptions increased, symbolic systems became increasingly fragile, requiring continual maintenance while struggling to adapt beyond the specific domains for which they had been designed.</p><p>Reasoning alone, it seemed, was not enough.</p><h3>Deep Learning Solved the Opposite Problem</h3><p>The rise of deep learning fundamentally altered the direction of artificial intelligence.</p><p>Rather than teaching computers how to reason explicitly, researchers began allowing neural networks to discover statistical relationships directly from enormous quantities of data.</p><p>This represented more than an incremental improvement.</p><p>It was a complete philosophical reversal.</p><p>Instead of constructing intelligence from predefined logical rules, neural networks learned through experience. Patterns emerged naturally during training without programmers explicitly specifying grammatical rules, artistic principles, or semantic relationships.</p><p>Unexpectedly, this approach succeeded spectacularly.</p><p>Modern language models can write essays, generate convincing dialogue, produce realistic artwork, compose music, translate between languages, summarise research papers, and explain highly technical concepts- all without ever being explicitly programmed to understand these tasks.</p><p>Perhaps the greatest surprise was not that machines became intelligent.</p><p>It was the <em>kind</em> of intelligence they developed.</p><p>For much of the twentieth century, both researchers and popular culture assumed that logic would prove relatively easy for computers, while creativity, humour, storytelling, and artistic expression would remain uniquely human.</p><p>Science fiction reflected this assumption repeatedly.</p><p>Machines were portrayed as brilliant calculators capable of flawless logic, yet fundamentally incapable of imagination or emotion. Characters such as Data from <em>Star Trek</em> or Sonny from <em>I, Robot</em> were memorable precisely because they struggled with creativity rather than computation.</p><p>Reality unfolded almost in reverse.</p><p>Today’s neural networks routinely generate paintings, novels, music, software, and persuasive writing that would have seemed impossible only a decade ago.</p><p>Yet these same systems occasionally hesitate over elementary logical comparisons, produce internally inconsistent reasoning, or confidently fabricate information.</p><p>This inversion has forced researchers to reconsider long-held assumptions about intelligence itself.</p><p>Rather than resembling perfect logicians, modern language models increasingly resemble extraordinarily powerful intuition engines.</p><p>They excel at recognising patterns across vast amounts of information. They identify relationships, infer context, complete incomplete ideas, and generate remarkably plausible responses by drawing upon statistical regularities acquired during training.</p><p>In other words, they function astonishingly well as something analogous to <strong>System 1</strong>.</p><p>Their weaknesses are equally revealing.</p><p>Like human intuition operating without careful reflection, they occasionally jump to conclusions, overgeneralise familiar patterns, or produce answers that <em>feel</em> convincing despite containing subtle logical errors.</p><p>Seen through this lens, many of the behaviours we describe as “hallucinations” or reasoning failures appear less mysterious.</p><p>They are not necessarily evidence that language models lack intelligence altogether.</p><p>Rather, they may indicate that we have become extraordinarily good at building one half of cognition while only beginning to understand the other.</p><p>Recent advances in reasoning-oriented models- where systems deliberately spend additional computational effort verifying intermediate steps before producing final answers- suggest that researchers are already attempting to bridge this gap. These techniques do not fundamentally replace the intuitive capabilities of large language models. Instead, they encourage models to behave more as deliberate reasoners layered upon an already powerful intuitive foundation.</p><p>Whether this approach ultimately leads towards human-level intelligence remains impossible to predict.</p><p>Perhaps entirely new neural architectures will eventually integrate intuition and reasoning seamlessly. Perhaps symbolic reasoning will return in modern hybrid forms. Or perhaps future AI systems will achieve robust intelligence through entirely different mechanisms that have yet to be discovered.</p><p>Whatever path emerges, one lesson appears increasingly clear.</p><p>If today’s language models represent an extraordinarily capable engine of intuition rather than a complete model of human cognition, then simply making them larger may never have been the entire solution.</p><p>The future of artificial intelligence may therefore depend not upon scaling intuition indefinitely, but upon discovering how intuition can become one component within a far richer system of intelligence.</p><h3>Why Large Language Models May Become Infrastructure Rather Than the End Goal</h3><p>If history teaches us anything about transformative technologies, it is that the first breakthrough is rarely the final product.</p><p>The inventions we celebrate today often began life as remarkably simple building blocks whose true significance only became apparent years later. The technologies that ultimately reshaped society did not do so because they were individually capable of solving every problem. Rather, they provided a reliable foundation upon which countless other innovations could be constructed.</p><p>Artificial intelligence may now be approaching a similar moment.</p><p>Much of today’s conversation revolves around whether the next frontier model will outperform the previous one, whether parameter counts will continue growing, or whether training runs will require even greater computational resources. These are important questions, but they may not be the most consequential ones.</p><p>A more interesting possibility is that large language models have already become something far more valuable than a destination.</p><p>They may have become infrastructure.</p><h3>Every Technological Revolution Begins With a Primitive</h3><p>Looking backwards, remarkable technological revolutions often share an unexpected characteristic.</p><p>They begin with a surprisingly modest primitive.</p><p>Consider modern computing.</p><p>Everything from smartphones and autonomous vehicles to supercomputers and cloud infrastructure ultimately traces its origins to a single electronic component: the transistor.</p><p>By itself, a transistor is remarkably unremarkable. It performs one simple task exceptionally well- it acts as an electrical switch, reliably alternating between conducting and insulating current. No individual transistor can browse the internet, render three-dimensional graphics, or execute sophisticated algorithms.</p><p>Its importance lies elsewhere.</p><p>Because transistors are reliable, inexpensive, and easily replicated, engineers learned to combine billions of them into increasingly sophisticated architectures. Progress did not require inventing a fundamentally different transistor every decade. Instead, innovation shifted towards designing better processors, memory systems, operating systems, and software using the same underlying building block.</p><p>The internet followed a remarkably similar trajectory.</p><p>Its foundational innovation was not social media, search engines, video streaming, or cloud computing.</p><p>It was the packet.</p><p>Packet switching introduced a deceptively simple concept: divide information into small, standardised units capable of travelling independently through a network before being reassembled at their destination.</p><p>Once this primitive existed, entirely new layers of technology emerged.</p><p>Communication protocols, distributed systems, web browsers, online commerce, video conferencing, and modern cloud infrastructure all evolved from repeatedly combining the same basic mechanism in increasingly sophisticated ways.</p><p>The revolution occurred not because packets became dramatically more intelligent.</p><p>It occurred because engineers discovered what could be built upon them.</p><h3>Large Language Models May Represent a Similar Primitive</h3><p>Artificial intelligence increasingly appears to be following the same historical pattern.</p><p>Rather than viewing modern language models as incomplete attempts at artificial general intelligence, it may be more useful to view them as the first genuinely universal cognitive primitive.</p><p>Unlike traditional software, which must be explicitly programmed for each task, large language models perform something remarkably general.</p><p>They absorb enormous quantities of unstructured information, identify statistical relationships across diverse domains, and generate plausible responses for problems they were never specifically designed to solve.</p><p>Whether processing natural language, computer code, medical literature, financial reports, images, or audio, the underlying mechanism remains fundamentally the same.</p><p>The model predicts meaningful patterns.</p><p>That capability may sound deceptively simple.</p><p>Yet it represents something fundamentally different from previous generations of software.</p><p>For decades, computers excelled only when problems could be described through precise logical rules. Every capability required explicit instructions written by human programmers.</p><p>Language models invert that relationship.</p><p>Rather than following carefully constructed rules, they learn broad statistical representations that generalise across remarkably diverse tasks.</p><p>In doing so, they become less like specialised applications and more like general-purpose cognitive infrastructure.</p><p>One could think of them as offering something akin to <em>intuition as a service</em>.</p><p>They rapidly interpret ambiguous information, recognise patterns, synthesise knowledge, generate candidate solutions, and produce plausible first approximations across almost any domain involving unstructured data.</p><p>Individually, these responses may not always be perfect.</p><p>Collectively, however, they provide an extraordinarily flexible foundation upon which more sophisticated systems can be constructed.</p><h3>Infrastructure Is Valuable Because Others Build Upon It</h3><p>Infrastructure rarely represents the final user experience.</p><p>Few people think about electrical grids while charging their phones.</p><p>Most internet users never consider routing protocols when streaming a film.</p><p>Likewise, future users may interact with increasingly capable AI systems without ever directly engaging with a standalone language model.</p><p>Instead, language models may quietly operate beneath multiple layers of software, coordinating information, interpreting requests, generating plans, and supporting higher-level systems that appear considerably more intelligent than any individual model.</p><p>In other words, the future may belong less to isolated models and more to integrated systems.</p><p>Rather than asking a single model to solve every aspect of a complex problem, future AI platforms could distribute responsibilities across multiple specialised components.</p><p>One subsystem might retrieve relevant information from external databases.</p><p>Another could evaluate logical consistency.</p><p>A third might execute code, verify mathematical calculations, or simulate scientific experiments.</p><p>The language model would no longer function as the entire intelligence.</p><p>Instead, it would become the interface through which numerous specialised capabilities cooperate.</p><p>This transition mirrors the evolution of computing itself.</p><p>Modern operating systems coordinate processors, memory, storage, networking, security, and countless background services. No single component performs every task, yet together they create a coherent computing environment.</p><p>Artificial intelligence may be approaching a similar systems-level architecture.</p><h3>Intelligence Is Increasingly Becoming Modular</h3><p>This perspective also resolves one of the apparent contradictions surrounding modern language models.</p><p>Much of today’s criticism focuses on what these systems cannot reliably accomplish.</p><p>They occasionally hallucinate.</p><p>They struggle with long chains of logical deduction.</p><p>They remain susceptible to factual inaccuracies.</p><p>Viewed as independent digital minds, these shortcomings appear significant.</p><p>Viewed as infrastructure, they become considerably less problematic.</p><p>No engineer expects an electrical wire to function as a complete computer.</p><p>Likewise, no one expects a database to understand natural language or a graphics processor to perform legal analysis.</p><p>Infrastructure succeeds precisely because each component performs a specific role exceptionally well while relying upon complementary systems to compensate for its limitations.</p><p>Language models may eventually occupy a similar position.</p><p>Their extraordinary strength lies in understanding ambiguity, generating language, synthesising information, recognising patterns, and interacting naturally with humans.</p><p>Other systems can increasingly assume responsibility for formal reasoning, symbolic computation, factual verification, long-term memory, and deterministic execution.</p><p>Rather than replacing language models, these additions amplify their usefulness.</p><p>The result is an intelligence composed of multiple cooperating components rather than one monolithic neural network expected to solve every possible task.</p><h3>The Question Changes Entirely</h3><p>If this interpretation proves correct, then debates surrounding parameter counts or benchmark scores become somewhat less central.</p><p>The more important question becomes:</p><p><strong>What kinds of systems can be built using language models as foundational infrastructure?</strong></p><p>History suggests this is where the most profound innovations often emerge.</p><p>Few observers witnessing the first integrated circuits could have anticipated smartphones.</p><p>Likewise, early packet-switched networks revealed little about streaming media, cloud computing, or global social platforms.</p><p>The enabling technology mattered enormously.</p><p>The applications built upon it mattered even more.</p><p>Large language models may currently occupy precisely that stage of technological development.</p><p>They are unlikely to represent the final expression of artificial intelligence.</p><p>Instead, they may prove to be the universal cognitive substrate upon which entirely new categories of intelligent systems are constructed.</p><p>If the scaling era eventually slows, that need not signify the end of progress.</p><p>It may simply mark the point at which artificial intelligence begins its transition from individual models towards something considerably more powerful: an ecosystem of interconnected cognitive systems, all built upon a surprisingly simple yet extraordinarily capable foundation.</p><h3>Small Models, Efficient Inference, Multi-Agent Systems, and the Future of AI</h3><p>If large language models eventually become the foundation rather than the destination of artificial intelligence, then the next question becomes obvious.</p><p>What does the architecture built on top of that foundation actually look like?</p><p>For much of the past few years, the industry’s answer has been remarkably consistent: build larger models. More parameters, more training data, more compute, and larger clusters were expected to produce increasingly capable systems. That strategy delivered extraordinary results, but if improvements from scaling alone begin to slow, progress will inevitably shift elsewhere.</p><p>History suggests that this transition would not represent stagnation.</p><p>Instead, it would signal a change in where innovation occurs.</p><p>Rather than concentrating almost exclusively on training ever-larger foundation models, researchers and engineers may increasingly focus on making intelligence more efficient, more specialised, and more collaborative. The future may therefore be defined not by one enormous artificial intelligence attempting to perform every conceivable task, but by networks of intelligent systems working together, each optimised for a particular role.</p><p>Three developments, in particular, appear likely to shape that future: specialised small models, dramatically more efficient inference, and autonomous AI agents.</p><h3>Intelligence Does Not Always Need to Be General</h3><p>One of the assumptions inherited from the race to build frontier models is that larger automatically means better.</p><p>In practice, however, many real-world problems are surprisingly narrow.</p><p>A physician interpreting radiological scans does not require encyclopaedic knowledge of medieval literature. Likewise, an AI responsible for detecting financial fraud does not necessarily benefit from mastering creative writing or translating between dozens of languages.</p><p>Many applications demand expertise rather than universality.</p><p>This is where small language models become increasingly compelling.</p><p>Unlike frontier models containing hundreds of billions of parameters, smaller models can be trained or fine-tuned for highly specific domains while remaining dramatically faster and less computationally expensive to operate.</p><p>Because they are focused on a narrower body of knowledge, they frequently achieve performance comparable to, or occasionally exceeding, that of much larger general-purpose models within their area of specialisation.</p><p>Equally important is accessibility.</p><p>Compact models are increasingly capable of running directly on consumer hardware, from laptops to smartphones and even embedded devices. Instead of transmitting sensitive information to distant cloud servers, inference can occur locally, improving privacy while reducing latency and infrastructure costs.</p><p>This shift mirrors a broader trend already visible across computing.</p><p>Cloud infrastructure continues expanding, yet edge computing has simultaneously become increasingly important. Rather than centralising every computational task, intelligent systems distribute workloads according to where they can be executed most effectively.</p><p>Artificial intelligence is likely to evolve in much the same way.</p><p>Large foundation models will continue providing broad capabilities, while smaller domain-specific models increasingly handle specialised tasks closer to where information is generated.</p><h3>Efficiency May Become the New Scaling Law</h3><p>Training a frontier model captures public attention.</p><p>Running one captures the electricity bill.</p><p>Inference- the process of using an already-trained model to generate outputs- is rapidly becoming one of the most important engineering challenges in artificial intelligence.</p><p>As AI systems become integrated into search engines, productivity software, healthcare, robotics, education, and scientific research, billions of inferences will eventually occur every day. Even modest improvements in computational efficiency therefore translate into enormous reductions in operational cost, energy consumption, and environmental impact.</p><p>The next decade may therefore be characterised less by scaling training and more by scaling inference.</p><p>This transformation is already underway.</p><p>Hardware manufacturers continue developing increasingly specialised AI accelerators optimised specifically for neural network inference rather than general-purpose computation. Custom silicon, improved memory hierarchies, and dedicated tensor-processing architectures significantly reduce the computational cost associated with deploying language models at scale.</p><p>Algorithmic improvements are equally important.</p><p>Techniques such as speculative decoding allow smaller auxiliary models to predict candidate outputs before larger models verify them, substantially accelerating text generation while maintaining quality.</p><p>Architectural innovations, particularly <strong>Mixture-of-Experts (MoE)</strong> models, represent another significant development.</p><p>Rather than activating every parameter during every inference, MoE architectures selectively engage only those expert networks most relevant to the current task. This dramatically reduces computational requirements while preserving the expressive capacity of much larger models.</p><p>Quantisation, pruning, distillation, and increasingly sophisticated compiler optimisations continue pushing efficiency even further.</p><p>Collectively, these advances suggest that future progress may depend as much upon making intelligence economically scalable as making it intellectually stronger.</p><p>After all, prohibitively expensive intelligence remains difficult to deploy widely, regardless of how impressive its capabilities may be.</p><h3>From Models That Respond to Systems That Act</h3><p>Perhaps the most profound transition, however, lies beyond models altogether.</p><p>Today’s language models are fundamentally reactive.</p><p>A user submits a prompt.</p><p>The model generates a response.</p><p>The interaction ends.</p><p>Increasingly, researchers are exploring systems capable of something considerably more ambitious.</p><p>Rather than merely answering questions, future AI systems may independently pursue objectives.</p><p>This distinction gives rise to the concept of the <strong>AI agent</strong>.</p><p>Although definitions continue evolving, an agent can broadly be understood as an intelligent system capable of understanding goals, planning sequences of actions, interacting with external tools, evaluating outcomes, and adapting its behaviour as circumstances change.</p><p>Instead of generating a single answer, an agent performs work.</p><p>Imagine requesting an AI to organise an international conference.</p><p>Rather than simply providing recommendations, an agent could search for venues, compare accommodation options, coordinate schedules, prepare budgets, communicate with suppliers, monitor progress, and adjust plans as new information becomes available.</p><p>The language model becomes only one component within a much larger decision-making architecture.</p><p>Memory systems preserve long-term context.</p><p>Reasoning modules evaluate competing strategies.</p><p>Planning algorithms organise complex workflows.</p><p>External software tools execute concrete actions.</p><p>The result resembles less a chatbot and more a digital collaborator.</p><h3>Intelligence Through Cooperation</h3><p>Human civilisation is not built upon individuals who each possess every possible skill.</p><p>It is built upon collaboration.</p><p>Architects design buildings.</p><p>Engineers construct them.</p><p>Doctors practise medicine.</p><p>Scientists conduct research.</p><p>Each specialist contributes expertise that others lack.</p><p>Collectively, the system becomes far more capable than any individual participant.</p><p>Artificial intelligence may ultimately evolve in the same direction.</p><p>Instead of relying upon one monolithic model expected to perform every conceivable task, future systems may consist of thousands- or eventually millions- of specialised agents cooperating within larger computational ecosystems.</p><p>One agent may retrieve information.</p><p>Another may verify factual accuracy.</p><p>A third could perform mathematical reasoning.</p><p>Others might manage scheduling, financial analysis, programming, scientific modelling, robotics, or communication.</p><p>Complex objectives would emerge through coordination rather than individual capability.</p><p>Recent developments already hint at this trajectory.</p><p>Multi-agent research increasingly demonstrates that specialised models collaborating through structured communication often outperform isolated models attempting to solve identical problems independently.</p><p>The intelligence emerges not solely from individual capability but from the organisation of the system itself.</p><p>This observation reflects an important principle found repeatedly throughout biology, economics, and engineering.</p><p>Complex systems frequently derive their power less from extraordinary components than from extraordinarily effective coordination between relatively simple ones.</p><p>Artificial intelligence may ultimately follow the same pattern.</p><h3>Looking Beyond the Scaling Debate</h3><p>Whether today’s large language models have truly reached a scaling wall remains uncertain.</p><p>Perhaps entirely new neural architectures will unlock another decade of exponential progress. Perhaps reinforcement learning, synthetic data, or multimodal training will reveal capabilities that remain difficult to imagine today.</p><p>History advises caution whenever we attempt to predict technological limits.</p><p>Yet even if frontier models continue improving, the broader direction of artificial intelligence already appears to be expanding beyond the pursuit of ever-larger networks.</p><p>The future is becoming increasingly architectural rather than purely algorithmic.</p><p>Instead of asking how to build one infinitely capable intelligence, researchers are beginning to ask how many different forms of intelligence can cooperate within a single system.</p><p>That subtle shift may ultimately prove more significant than another trillion parameters.</p><p>Large language models have given us something genuinely unprecedented: a remarkably general engine for understanding and generating information. Whether they represent the first step towards artificial general intelligence or simply the cognitive equivalent of the transistor remains impossible to know.</p><p>What seems increasingly plausible, however, is that their greatest contribution will not be replacing every other form of intelligence.</p><p>It will enable entirely new ways of organising intelligence itself.</p><p>If that future arrives, artificial intelligence will not be defined by one model answering every question.</p><p>It will be defined by millions of specialised systems working quietly together- efficiently, collaboratively, and often invisibly- to solve problems far beyond the capabilities of any single model alone.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=f1b4ae2a464c" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Why Orbital Computing Could Redefine the Future of AI Infrastructure]]></title>
            <link>https://medium.com/@kavyarathi1102/why-orbital-computing-could-redefine-the-future-of-ai-infrastructure-74cb89c57950?source=rss-e9b9ce957cf4------2</link>
            <guid isPermaLink="false">https://medium.com/p/74cb89c57950</guid>
            <category><![CDATA[engineering]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[space-technology]]></category>
            <category><![CDATA[computer-science]]></category>
            <category><![CDATA[space-exploration]]></category>
            <dc:creator><![CDATA[Kavya Rathi]]></dc:creator>
            <pubDate>Mon, 29 Jun 2026 17:22:53 GMT</pubDate>
            <atom:updated>2026-06-29T17:22:53.519Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*IVG_FHpSBL4KzOyZ" /><figcaption>Photo by <a href="https://unsplash.com/@nasa?utm_source=medium&amp;utm_medium=referral">NASA</a> on <a href="https://unsplash.com/?utm_source=medium&amp;utm_medium=referral">Unsplash</a></figcaption></figure><p>Every generation inherits a technological assumption that eventually stops being true.</p><p>For decades, we have assumed that computation belongs on Earth. Data centres have always been anchored to the same fundamentals: abundant electricity, readily available cooling, human maintenance, and a planet whose atmosphere quietly solves countless engineering problems without us ever thinking about it. As artificial intelligence accelerates, however, those assumptions are beginning to look less permanent than they once did.</p><p>The conversation surrounding orbital computing has grown considerably over the past year, fuelled by rapid advances in launch capabilities and a renewed interest in space-based infrastructure. What once belonged almost exclusively to science fiction is now becoming an engineering discussion. The question is no longer whether computers <em>can</em> operate in space -they already do aboard satellites and spacecraft -but whether entire high-performance computing clusters could one day exist in orbit, processing some of the world’s most demanding workloads.</p><p>The appeal is understandable. Artificial intelligence is driving an unprecedented demand for computational power, while terrestrial data centres face increasing pressure from limited electrical grids, rising construction costs, water consumption, and environmental constraints. Launch costs continue to decline, satellite manufacturing has become increasingly sophisticated, and reusable rockets have fundamentally changed the economics of reaching low Earth orbit. Together, these developments invite a compelling question: could space eventually become an extension of Earth’s digital infrastructure?</p><p>The idea is bold, but transforming it into reality demands far more than launching conventional servers aboard a rocket. Virtually every design principle underpinning modern computing has been developed for conditions that simply do not exist beyond our atmosphere. Gravity disappears, convection ceases to function, radiation intensifies, temperatures fluctuate dramatically, and physical maintenance becomes almost impossible. In orbit, engineering is dictated not by convenience, but by physics.</p><p>Building a successful orbital data centre therefore requires rethinking the entire computing stack -from semiconductor packaging to thermal management, power generation, communication systems, and long-term reliability. Every subsystem must be redesigned to function in an environment where familiar engineering shortcuts no longer apply.</p><p>Perhaps nowhere is this challenge more evident than in managing heat.</p><h3>Heat: The Constraint That Changes Everything</h3><p>Modern processors convert an extraordinary amount of electrical energy into heat. A single high-performance GPU may consume several hundred watts under sustained workloads, and advanced AI clusters routinely dissipate megawatts of thermal energy. On Earth, removing this heat is comparatively straightforward. Fans circulate air across heatsinks, chilled water carries energy away from server racks, and cooling towers ultimately release it into the atmosphere.</p><p>Space offers none of these luxuries.</p><p>A vacuum contains virtually no matter capable of transporting thermal energy through convection. There is no surrounding air to absorb heat, no breeze to cool exposed surfaces, and no atmospheric circulation to assist engineers. Every joule generated inside a processor must be carefully transported and ultimately emitted as infrared radiation into space itself.</p><p>This seemingly simple difference transforms thermal engineering into one of the defining problems of orbital computing.</p><h3>The First Challenge: Removing Heat From the Processor</h3><p>Heat begins its journey inside the silicon.</p><p>Transistors switching billions of times every second generate highly concentrated thermal hotspots that must be transferred immediately into a cooling system. Even microscopic inefficiencies at this stage can produce dramatic temperature increases capable of reducing performance or permanently damaging the chip.</p><p>Terrestrial electronics rely heavily on thermal interface materials- specialised compounds placed between processors and heatsinks to eliminate microscopic air gaps and maximise thermal conductivity. These materials perform exceptionally well within Earth’s atmosphere, but orbit introduces an unexpected complication.</p><p>Many conventional thermal compounds slowly release trapped volatile molecules when exposed to vacuum, a phenomenon known as <strong>outgassing</strong>. As these materials degrade, tiny voids begin to develop between the processor and its cooling interface. Because air is an exceptionally poor thermal conductor, even microscopic pockets act as insulation, dramatically increasing thermal resistance and trapping heat precisely where efficient cooling is most critical.</p><p>Rather than treating thermal interfaces as consumable materials, orbital systems are likely to depend on far more stable alternatives. Liquid-metal interfaces offer thermal conductivities significantly higher than traditional pastes while remaining resistant to vacuum-induced degradation. Likewise, synthetic diamond -one of the most thermally conductive materials known -can spread heat across a much larger surface before it reaches the primary cooling system.</p><p>Packaging technologies are also evolving. Glass-core substrates, increasingly explored for next-generation semiconductor packaging, provide superior dimensional stability and improved management of thermal expansion compared with conventional organic substrates. As processors become increasingly dense, the architecture surrounding the chip becomes almost as important as the silicon itself.</p><h3>The Second Challenge: Moving Heat Across the Spacecraft</h3><p>Extracting heat from the processor is only the beginning.</p><p>That energy must still travel across the spacecraft to external cooling structures capable of rejecting it into space. On Earth, copper pipes carrying liquid coolant accomplish this with remarkable efficiency. Unfortunately, copper is heavy, and every additional kilogram launched into orbit carries a substantial financial penalty.</p><p>Weight alone is not the only obstacle.</p><p>Microgravity fundamentally alters how boiling fluids behave.</p><p>Under terrestrial conditions, bubbles produced during boiling naturally rise away from heated surfaces because buoyancy separates liquid from vapour. This continuous movement maintains efficient heat transfer. In orbit, however, buoyancy effectively disappears. Vapour bubbles can remain attached to heated components, gradually forming insulating layers that prevent fresh coolant from reaching the surface.</p><p>This phenomenon, commonly associated with <strong>critical heat flux collapse</strong>, causes cooling performance to deteriorate abruptly. Temperatures may rise within milliseconds, potentially damaging expensive computing hardware long before protective software can intervene.</p><p>To overcome this, engineers are investigating <strong>pumped two-phase cooling systems</strong>, which exploit the enormous amount of energy absorbed during liquid-to-vapour phase transitions. Instead of relying on gravity, carefully engineered capillary structures and heat-controlled accumulators continuously guide liquid toward heated regions while removing vapour before it can accumulate. These systems move significantly more thermal energy per unit mass than conventional liquid cooling loops, making them particularly attractive for spacecraft where every kilogram matters.</p><p>The result is a cooling architecture that behaves less like a conventional plumbing system and more like a carefully orchestrated thermodynamic machine.</p><h3>The Final Challenge: Getting Rid of Heat Altogether</h3><p>Even after heat has been transported safely through the spacecraft, the most difficult problem remains.</p><p>Where does it go?</p><p>Unlike terrestrial data centres, spacecraft cannot simply exhaust warm air into the surrounding environment. Once thermal energy reaches the spacecraft’s exterior, the only remaining mechanism available is <strong>thermal radiation</strong>.</p><p>Every object with a temperature above absolute zero continuously emits electromagnetic radiation. The hotter an object becomes, the more energy it radiates. This relationship is described by the <strong>Stefan–Boltzmann law</strong>, which states that radiated power increases with the fourth power of absolute temperature.</p><p>For engineers, this creates an uncomfortable trade-off.</p><p>Running hardware at higher temperatures dramatically improves the efficiency of radiative cooling, potentially allowing much smaller radiator systems. Yet advanced AI accelerators are designed to operate within strict thermal limits. Temperatures approaching 85°C often trigger performance throttling, while prolonged overheating accelerates transistor degradation and shortens component lifespan.</p><p>Since processors cannot simply be allowed to operate indefinitely at much higher temperatures, radiator size becomes the remaining design variable.</p><p>Unfortunately, radiators consume valuable spacecraft mass, occupy large surface areas, and must contend not only with internally generated heat but also with solar radiation, reflected sunlight from Earth, and the planet’s own infrared emissions. Designing them therefore becomes a complex optimisation problem involving orbital geometry, spacecraft orientation, and mission duration.</p><p>Researchers are exploring several alternatives capable of fundamentally changing this equation.</p><p>One of the most ambitious concepts is the <strong>liquid droplet radiator</strong>. Rather than relying on large solid panels, these systems release controlled streams of microscopic liquid droplets into space. Each droplet radiates heat efficiently before being recollected and recirculated through the spacecraft. If successfully matured, such systems could dramatically reduce radiator mass while increasing overall heat rejection capacity.</p><p>Progress in semiconductor materials may also contribute. Wide-bandgap materials such as <strong>gallium nitride (GaN)</strong> and <strong>silicon carbide (SiC)</strong> tolerate substantially higher operating temperatures than conventional silicon. Diamond-based electronic substrates are similarly attracting attention because of their exceptional thermal conductivity. Together, these technologies could allow future processors to operate safely at temperatures that make radiative cooling considerably more effective.</p><p>Thermal engineering is therefore not simply one challenge among many- it is the foundation upon which orbital computing depends. Before questions of artificial intelligence, networking, or economics can be answered, engineers must first solve the far more fundamental problem of ensuring that a computer in the vacuum of space can continue to keep itself cool. In many respects, the future of orbital computing will be determined not by faster processors, but by better ways of moving heat.</p><h3>The Unseen Threat: Radiation and the Challenge of Reliability</h3><p>Heat may be the most immediate engineering challenge in orbit, but radiation is arguably the most unforgiving. Unlike thermal problems, which can be managed through careful design, radiation cannot be eliminated. Every spacecraft operating beyond Earth’s atmosphere exists within an environment saturated with high-energy particles capable of altering, degrading, and sometimes permanently destroying modern electronics.</p><p>Here on Earth, our atmosphere and magnetic field act as remarkably effective shields. They absorb or deflect the overwhelming majority of energetic particles arriving from deep space or emitted by the Sun. Orbiting spacecraft enjoy no such protection. Every processor, memory module, and transistor must continuously withstand an invisible bombardment that terrestrial data centres never experience.</p><p>As semiconductor manufacturing continues pushing toward smaller process nodes- 5 nm, 3 nm, and beyond- the problem only becomes more severe. Modern transistors store information using extraordinarily small electrical charges. While this enables remarkable computational performance, it also means increasingly tiny amounts of energy are required to accidentally alter a circuit’s behaviour.</p><p>Ironically, the same technological advances that make AI accelerators more powerful also make them more vulnerable.</p><h3>Instantaneous Failures: When a Single Particle Is Enough</h3><p>Not all radiation damage is gradual.</p><p>Sometimes, a single energetic particle passing through a semiconductor can immediately alter the state of a transistor. Known as a <strong>Single-Event Upset (SEU)</strong>, this phenomenon occurs when sufficient charge is deposited within a logic circuit to change a stored binary value.</p><p>The consequences extend far beyond a simple bit flip.</p><p>In large AI systems, corrupted memory may propagate unnoticed through billions of mathematical operations. A scientific simulation may produce subtly incorrect outputs. A machine-learning model might return inaccurate predictions without generating any obvious error message. These so-called <em>silent data corruptions</em> are especially concerning because they often appear indistinguishable from legitimate computational results.</p><p>Even more destructive are <strong>Single-Event Latchups (SELs)</strong>.</p><p>Instead of merely changing stored information, certain radiation events activate unintended conductive pathways inside CMOS circuitry, effectively creating an electrical short circuit between power and ground. Current rises rapidly, temperatures escalate, and unless power is removed almost instantaneously, permanent physical damage to the silicon can occur.</p><p>Unlike ordinary software crashes, these failures are not resolved with a system reboot. They represent genuine hardware destruction.</p><p>Designing around such risks requires multiple layers of protection rather than a single solution.</p><p>Future orbital computing platforms will likely combine commercial high-performance GPUs with radiation-tolerant supervisory processors responsible for continuously monitoring system health. These secondary controllers can verify computational outputs using <strong>Algorithm-Based Fault Tolerance (ABFT)</strong> techniques while simultaneously monitoring electrical behaviour for early signs of latchup.</p><p>Should abnormal current spikes be detected, the affected processor can be isolated and power-cycled within microseconds, often fast enough to prevent irreversible damage.</p><p>Memory protection also becomes substantially more sophisticated than in conventional servers. Continuous memory scrubbing, advanced error-correcting codes (ECC), redundant computation, and software-defined fault recovery collectively reduce the probability that isolated radiation events compromise large-scale AI workloads.</p><p>Rather than attempting to prevent every fault, modern spacecraft increasingly assume that faults are inevitable- and focus instead on detecting, correcting, and containing them before they cascade into catastrophic failures.</p><h3>The Slow March of Radiation Damage</h3><p>While instantaneous failures attract considerable attention, the cumulative effects of radiation often pose an even greater threat over long-duration missions.</p><p>Every energetic particle interacting with semiconductor materials deposits a tiny amount of charge within insulating layers surrounding transistors. Individually, these events appear insignificant. Over months and years, however, trapped charge gradually accumulates, permanently altering the electrical characteristics of the device.</p><p>This phenomenon is known as the <strong>Total Ionising Dose (TID)</strong> effect.</p><p>As threshold voltages slowly drift, transistors require different operating conditions to switch correctly. Leakage currents increase, signal timing becomes less predictable, and power consumption gradually rises. Processors effectively age at an accelerated rate, even when operating perfectly within their intended specifications.</p><p>Unlike mechanical wear, this degradation cannot be repaired.</p><p>Eventually, once accumulated radiation exceeds the device’s tolerance, reliable operation becomes impossible.</p><p>Mitigating long-term degradation begins at the materials level.</p><p><strong>Silicon-on-Insulator (SOI)</strong> technology offers significantly greater radiation tolerance than conventional bulk silicon by electrically isolating transistors from the underlying substrate. This architecture reduces charge accumulation and minimises parasitic electrical interactions caused by energetic particles.</p><p>Shielding also demands careful optimisation.</p><p>It might seem intuitive that simply adding thicker metal shielding would solve the problem. Unfortunately, high-energy cosmic particles striking dense metals often generate secondary radiation through interactions within the shielding itself. These secondary particles can sometimes prove almost as damaging as the original radiation.</p><p>For this reason, engineers increasingly favour <strong>graded multilayer shielding</strong>, alternating dense materials such as tantalum with lighter polymers. High-density layers slow incoming particles, while low-density materials absorb much of the resulting secondary radiation before it reaches sensitive electronics.</p><p>Protection, therefore, becomes an exercise in materials science as much as structural engineering.</p><h3>Power: Supplying an AI Data Centre Beyond Earth</h3><p>If heat determines whether orbital computing is physically possible, power determines whether it is practical.</p><p>Artificial intelligence has fundamentally altered the energy requirements of modern computing. Training frontier-scale models requires enormous computational clusters consuming tens or even hundreds of megawatts. Even inference workloads place sustained demands on accelerators operating continuously at high utilisation.</p><p>Replicating this environment in orbit presents challenges unlike those faced by conventional satellites.</p><p>Most spacecraft today consume anywhere from a few hundred watts to several kilowatts. An orbital AI data centre would require orders of magnitude more power while remaining constrained by launch mass, deployment complexity, and long-term reliability.</p><p>Generating electricity, therefore, becomes one of the defining engineering considerations.</p><h3>Rethinking Solar Power in Orbit</h3><p>Solar energy is the obvious candidate.</p><p>Outside Earth’s atmosphere, solar panels receive uninterrupted sunlight free from atmospheric absorption, providing a remarkably consistent energy source whenever the spacecraft remains illuminated.</p><p>Yet scaling this capability introduces significant complications.</p><p>Conventional photovoltaic cells remain fundamentally constrained by the <strong>Shockley-Queisser efficiency limit</strong>, restricting how much incoming sunlight can be converted into electrical energy. As computational demand increases, so too must the collecting area of the solar arrays.</p><p>Very quickly, power generation begins to dominate spacecraft design.</p><p>Extremely large solar wings increase structural complexity, deployment risk, and launch mass. They also create substantial aerodynamic drag in low Earth orbit. Although the atmosphere at these altitudes is extraordinarily thin, it is not entirely absent. Over time, atmospheric drag gradually lowers orbital altitude, requiring periodic propulsion burns to maintain position.</p><p>Larger solar arrays therefore increase both fuel consumption and operational costs throughout the satellite’s lifetime.</p><p>To improve efficiency without proportionally increasing size, engineers are turning toward <strong>multi-junction III-V photovoltaic cells</strong>.</p><p>Rather than relying upon a single semiconductor layer, these advanced cells stack multiple materials- including gallium indium phosphide (GaInP), gallium arsenide (GaAs), and germanium (Ge) -each optimised to absorb different portions of the solar spectrum. Laboratory efficiencies exceeding 40% have already demonstrated their superiority over conventional silicon technologies.</p><p>Further gains come from manufacturing innovations such as <strong>shingled-cell architectures</strong>, which minimise inactive surface area and improve power generation per kilogram launched into orbit.</p><p>In spacecraft engineering, every additional watt extracted from the same panel represents valuable mass saved elsewhere.</p><h3>The Problem of Darkness</h3><p>Even perfect solar cells cannot generate electricity without sunlight.</p><p>Low Earth orbit presents an unavoidable reality: approximately every ninety minutes, a spacecraft passes behind Earth and enters eclipse. During these periods, solar generation falls essentially to zero while onboard computers continue demanding uninterrupted power.</p><p>This transforms energy storage from a convenience into a mission-critical subsystem.</p><p>Unfortunately, conventional lithium-ion batteries perform poorly under the relentless charging and discharging cycles experienced in orbit. Satellites may encounter more than a dozen eclipse transitions every day, accumulating tens of thousands of cycles throughout their operational lifetime.</p><p>Repeated cycling gradually promotes <strong>lithium plating</strong>, where metallic lithium deposits form within battery cells during charging. These deposits reduce capacity, increase internal resistance, and can eventually create dangerous internal short circuits.</p><p>For orbital computing platforms expected to operate continuously for years, battery longevity becomes just as important as battery capacity.</p><p>One promising alternative is <strong>Lithium Titanate (LTO)</strong> chemistry.</p><p>Although LTO batteries store less energy per unit mass than conventional lithium-ion cells, they tolerate extraordinarily high cycle counts- often exceeding 70,000 deep charge-discharge cycles- while maintaining excellent thermal stability across extreme temperatures. Their resistance to rapid degradation makes them particularly well suited for spacecraft enduring frequent eclipse transitions.</p><p>Researchers are also exploring <strong>solid-state batteries</strong>, which replace flammable liquid electrolytes with solid materials. Besides improving safety, these systems eliminate concerns surrounding electrolyte outgassing in vacuum and may ultimately offer higher energy density alongside improved long-term reliability.</p><p>For orbital AI infrastructure, energy storage is not merely about surviving periods of darkness. It is about ensuring that computation remains uninterrupted despite an environment where sunrise and sunset occur sixteen times every day.</p><p>Power, much like thermal management, ultimately becomes a balancing act between physics, materials science, and engineering practicality. Before a single neural network can process data in orbit, engineers must first solve the far more fundamental question of how to generate, store, and deliver vast quantities of reliable electrical energy hundreds of kilometres above Earth.</p><h3>Connecting Orbit to Earth: The Communications Challenge</h3><p>A data centre derives little value from computation alone. Its true purpose lies in transforming data into useful information and delivering those results to users. Whether training large language models, processing satellite imagery, or running scientific simulations, computation is only half the equation. The remaining challenge is ensuring that enormous quantities of data can move reliably between Earth and orbit.</p><p>This is where orbital computing encounters another fundamental constraint.</p><p>Unlike terrestrial infrastructure, where fibre-optic cables routinely carry terabits of data every second, spacecraft cannot rely on physical connections. Every byte must be transmitted wirelessly across hundreds or sometimes thousands of kilometres. As AI workloads continue to expand, the communication system increasingly risks becoming the limiting factor rather than the processors themselves.</p><p>In essence, there is little advantage in building an exceptionally powerful computer if its outputs cannot return to Earth quickly enough to be useful.</p><h3>Moving Beyond Traditional Radio</h3><p>Historically, satellites have communicated using radio frequency (RF) transmissions. These systems are well understood, highly reliable, and capable of operating through most weather conditions. However, they were never designed with modern AI infrastructure in mind.</p><p>Radio frequencies occupy a finite portion of the electromagnetic spectrum, meaning bandwidth is inherently limited. As more satellites enter orbit and global demand for communication increases, RF channels become increasingly congested. Even with advanced modulation techniques, sustaining the enormous data rates expected from orbital AI clusters would prove extremely challenging.</p><p>For this reason, engineers are increasingly looking towards <strong>Free Space Optical (FSO)</strong> communication.</p><p>Rather than transmitting radio waves, FSO systems use tightly focused infrared laser beams- typically operating near a wavelength of 1550 nanometres- to transfer information. Optical frequencies provide dramatically greater bandwidth than conventional RF systems, allowing vastly larger volumes of data to be transmitted while reducing interference from neighbouring spacecraft.</p><p>Conceptually, laser communication functions much like an invisible fibre-optic cable stretching between satellites and ground stations- except there is no physical fibre connecting either end.</p><p>The benefits are substantial.</p><p>Laser beams spread far less than radio waves over long distances, allowing more energy to reach the receiver while reducing the risk of interception or interference. Higher bandwidth also makes optical communication particularly attractive for transferring the immense datasets generated by machine learning workloads.</p><p>Yet these advantages come with extraordinary engineering demands.</p><h3>Precision at an Almost Unimaginable Scale</h3><p>Maintaining an optical communication link requires astonishing accuracy.</p><p>Unlike radio transmissions, which naturally spread over wide angles, laser beams remain extremely narrow. Even microscopic pointing errors become significant across orbital distances.</p><p>A deviation of only a few microradians- roughly equivalent to the angle subtended by a coin viewed from several kilometres away- can cause a laser beam to miss its intended receiver entirely.</p><p>Unfortunately, spacecraft are never perfectly stable.</p><p>Reaction wheels continuously adjust spacecraft orientation, internal equipment generates small vibrations, and thermal expansion causes subtle structural movements as satellites repeatedly transition between sunlight and shadow. Individually, these disturbances appear insignificant. Collectively, they introduce enough motion to disrupt optical links unless actively compensated.</p><p>To overcome this, modern laser communication terminals incorporate <strong>Fast Steering Mirrors (FSMs)</strong> capable of making thousands of microscopic corrections every second. Driven by highly responsive piezoelectric actuators, these mirrors continuously redirect the outgoing beam, cancelling spacecraft jitter before it affects communication.</p><p>Maintaining the beam’s strength across vast distances presents another challenge.</p><p>Rather than repeatedly converting optical signals into electrical ones for amplification, engineers employ <strong>Erbium-Doped Fibre Amplifiers (EDFAs)</strong>, which strengthen infrared laser signals directly within optical fibres. Eliminating unnecessary electronic conversions reduces signal degradation while enabling communication over thousands of kilometres with remarkable efficiency.</p><p>Together, these technologies transform what might otherwise be an impossibly delicate communication system into one capable of supporting high-bandwidth orbital networks.</p><h3>The Final Obstacle: Earth’s Atmosphere</h3><p>Ironically, the most difficult portion of an optical communication link is often the final few kilometres.</p><p>Space itself is almost an ideal transmission medium. There is no atmosphere to scatter light, no weather, and virtually no interference. Problems begin only when laser beams encounter Earth’s atmosphere.</p><p>The atmosphere is far from uniform.</p><p>Rising warm air, falling cool air, and constantly shifting pockets of varying density alter the refractive index of the air through which the beam travels. As the laser passes through these turbulent layers, it bends slightly in unpredictable ways, causing fluctuations in intensity and position by the time it reaches the ground.</p><p>This effect, known as <strong>scintillation</strong>, resembles the twinkling of stars observed from Earth. For communication systems, however, these fluctuations reduce signal quality and increase transmission errors.</p><p>Clouds introduce an even more severe limitation.</p><p>Water droplets and atmospheric aerosols scatter infrared light through processes such as <strong>Mie scattering</strong>, often weakening optical signals dramatically. Thick cloud cover can block laser communication altogether, temporarily severing the connection between spacecraft and ground station.</p><p>Because weather cannot be controlled, engineers instead design networks capable of adapting to it.</p><p><strong>Adaptive optics</strong> systems monitor incoming wavefront distortions in real time using reference signals and rapidly reshape deformable mirrors to compensate for atmospheric turbulence before the data are decoded.</p><p>At a broader network level, operators increasingly rely on <strong>ground site diversity</strong>. Rather than depending upon a single receiving station, multiple geographically separated facilities operate simultaneously. Should poor weather affect one location, communication can simply be redirected to another station experiencing clear skies.</p><p>In cases where optical links remain temporarily unavailable, lower-bandwidth Ka-band radio systems may serve as backup channels, ensuring uninterrupted connectivity even under unfavourable atmospheric conditions.</p><p>The future of orbital computing therefore depends not only upon faster processors but equally upon communication networks capable of transporting unprecedented quantities of information between Earth and space.</p><h3>Engineering for an Environment Unlike Any Other</h3><p>Long before a spacecraft begins processing AI workloads, it must first survive the journey into orbit.</p><p>Unlike terrestrial servers, which travel from factory to data centre with relatively little mechanical stress, orbital computing hardware experiences one of the most violent transportation environments engineers routinely encounter.</p><p>Every design decision must account for forces rarely considered in conventional computing.</p><h3>Surviving Launch</h3><p>Rocket launches subject spacecraft to sustained acceleration, intense acoustic energy, random vibration, and violent shocks generated during stage separation.</p><p>Throughout ascent, structural loads may reach twenty times Earth’s gravitational acceleration. Simultaneously, engines generate broadband vibrations capable of exciting resonant frequencies throughout the spacecraft. Later, explosive separation mechanisms produce brief but extremely intense shock waves exceeding one thousand times gravitational acceleration.</p><p>Although these events last only minutes, they can permanently damage delicate electronics.</p><p>Silicon dies may fracture under excessive mechanical stress. Microscopic solder joints connecting processors to circuit boards can delaminate or crack. Precision optical components may lose alignment before the mission has even begun.</p><p>Preventing such failures begins long before any hardware is manufactured.</p><p>Engineers construct detailed <strong>Finite Element Analysis (FEA)</strong> models to simulate how every structural component responds to launch loads. Natural frequencies are carefully adjusted to avoid resonance with expected rocket vibrations, while passive isolation systems absorb high-frequency energy before it reaches sensitive electronics.</p><p>By the time hardware is integrated into a launch vehicle, it has often survived months of qualification testing designed to replicate conditions even more severe than those expected during flight.</p><h3>Living Through Sixteen Sunrises Every Day</h3><p>Reaching orbit marks the beginning- not the end- of environmental challenges.</p><p>Spacecraft in low Earth orbit repeatedly transition between direct sunlight and Earth’s shadow, experiencing temperature swings that may exceed 200°C over the course of a single orbit.</p><p>Unlike terrestrial structures, these thermal cycles occur approximately sixteen times every day.</p><p>The problem is not simply that temperatures become extremely hot or cold.</p><p>Different materials expand and contract at different rates.</p><p>Silicon, copper, aluminium, ceramics, composite structures, and printed circuit boards each possess unique <strong>coefficients of thermal expansion</strong>. As temperatures fluctuate, neighbouring materials attempt to change dimensions by different amounts, creating continuous mechanical stress wherever they are bonded together.</p><p>Initially, these stresses remain harmless.</p><p>Over thousands of thermal cycles, however, microscopic fatigue cracks begin developing within solder joints, adhesive layers, and electrical interconnections. What starts as an invisible defect gradually grows until an electrical connection fails.</p><p>For spacecraft expected to remain operational for many years, thermomechanical fatigue becomes one of the principal drivers of long-term reliability.</p><p>Engineers therefore devote considerable attention to selecting materials whose thermal behaviour remains compatible over prolonged cycling.</p><p>Highly conductive <strong>Annealed Pyrolytic Graphite (APG)</strong> has emerged as an attractive thermal bus material due to its exceptional ability to transport heat while adding relatively little mass. Likewise, specialised low-outgassing aerospace adhesives maintain structural integrity across repeated temperature extremes while minimising contamination within sensitive spacecraft systems.</p><p>Ultimately, orbital computing demands an engineering philosophy fundamentally different from terrestrial infrastructure. Every bolt, solder joint, adhesive, structural panel, and optical assembly must survive launch, vacuum, radiation, thermal cycling, and years of continuous operation without the possibility of routine maintenance.</p><p>Building computers for orbit is therefore not simply about making existing hardware smaller or lighter. It requires designing systems capable of thriving in one of the harshest environments humanity has ever attempted to engineer within.</p><h3>The Problem That Changes Everything: Maintenance in Orbit</h3><p>Modern data centres are designed with an assumption so deeply ingrained that it is almost invisible: hardware will eventually fail.</p><p>Processors degrade, memory modules develop faults, storage devices reach the end of their service lives, cooling pumps require replacement, and power supplies occasionally malfunction. None of this is particularly alarming because every component is accessible. Engineers replace defective hardware within minutes, workloads migrate seamlessly between servers, and routine maintenance forms part of everyday operations.</p><p>Orbital computing removes that assumption entirely.</p><p>Once a data centre has been launched into orbit, there is effectively no possibility of sending technicians to replace failed hardware. Every component must either continue functioning autonomously for years or be replaced through an entirely different approach.</p><p>This shifts reliability from being an operational consideration to becoming a fundamental design philosophy.</p><p>Rather than asking <em>how quickly a failed component can be repaired</em>, engineers must instead ask <em>how failures can be prevented, isolated, or recovered from without direct human intervention.</em></p><p>The distinction is subtle, yet it influences almost every aspect of spacecraft architecture.</p><h3>Designing for Failure Instead of Perfection</h3><p>Absolute reliability does not exist.</p><p>Even the most rigorously tested hardware will eventually experience failures, particularly after years of exposure to radiation, thermal cycling, and continuous operation. The objective, therefore, is not to eliminate every possible fault but to ensure that no single failure jeopardises the entire mission.</p><p>Future orbital computing platforms are expected to adopt highly modular architectures.</p><p>Instead of constructing monolithic spacecraft containing irreplaceable computing systems, engineers are increasingly exploring designs composed of interchangeable compute modules. Individual processing units could be electrically and mechanically isolated, allowing faulty sections to be disconnected while the remaining system continues operating with minimal interruption.</p><p>This philosophy mirrors the redundancy already employed in aviation and crewed spacecraft, where essential systems are duplicated to prevent isolated failures from escalating into mission-ending events.</p><p>The difference is that orbital data centres may eventually extend this concept to thousands of individual computing nodes.</p><h3>Robotic Servicing: A New Layer of Space Infrastructure</h3><p>While autonomous redundancy improves reliability, it cannot entirely replace maintenance.</p><p>As orbital infrastructure expands, robotic servicing is becoming an increasingly attractive alternative to human intervention.</p><p>Rather than launching entirely new satellites whenever a critical component fails, future servicing spacecraft could rendezvous with existing orbital data centres, identify defective modules using advanced computer vision systems, remove them with robotic manipulators, and install replacements.</p><p>Achieving this vision requires spacecraft designed specifically for robotic interaction.</p><p>Mechanical interfaces must be standardised. Electrical connectors need to tolerate repeated mating cycles in vacuum. Navigation systems must support autonomous docking with centimetre- or even millimetre- level precision.</p><p>Although still an emerging capability, early demonstrations of autonomous rendezvous and satellite servicing suggest that this concept is steadily moving from experimental missions towards practical infrastructure.</p><p>If successful, orbital servicing could fundamentally change the economics of space computing by extending operational lifetimes and reducing the frequency of complete satellite replacement.</p><h3>Technologies That Could Make Orbital Computing Practical</h3><p>Many of the engineering solutions discussed throughout this article already exist in some form. Others remain confined to laboratories or early experimental missions.</p><p>The success of orbital computing will ultimately depend upon how rapidly several independent technologies mature simultaneously.</p><p>Launch capability remains one of the most obvious factors.</p><p>The continued development of fully reusable heavy-lift launch vehicles has dramatically reduced the cost of placing payloads into orbit compared with previous decades. As launch prices continue declining, spacecraft can allocate more mass towards redundancy, shielding, thermal management, and power generation without rendering missions economically unviable.</p><p>Advances in semiconductor technology will prove equally significant.</p><p>Future processors built using wide-bandgap materials such as gallium nitride and silicon carbide promise higher operating temperatures, improved radiation tolerance, and greater electrical efficiency. Likewise, continued progress in advanced packaging, chiplet architectures, and three-dimensional integration may significantly reduce the energy consumed per computation.</p><p>Artificial intelligence itself may also become part of the solution.</p><p>Machine learning algorithms are increasingly capable of predicting hardware degradation before failures occur. Future orbital platforms may continuously analyse thermal behaviour, electrical signatures, vibration patterns, and radiation exposure to forecast failures days or even weeks before they become critical.</p><p>Instead of reacting to faults, spacecraft may eventually schedule preventive maintenance autonomously.</p><p>Communications infrastructure will likewise continue evolving.</p><p>Optical inter-satellite links are already beginning to replace traditional radio communications within large satellite constellations. As these laser networks expand, orbital computing facilities may exchange information directly with one another before transmitting aggregated results to Earth, reducing communication bottlenecks while improving overall network resilience.</p><p>Viewed individually, none of these technologies is sufficient.</p><p>Together, however, they begin to outline a plausible path towards large-scale computational infrastructure beyond Earth’s atmosphere.</p><h3>The Economic Question</h3><p>Even if every engineering challenge discussed throughout this article were solved, one fundamental question would remain.</p><p>Would orbital computing actually make economic sense?</p><p>Engineering feasibility and commercial viability are not always synonymous.</p><p>Constructing a terrestrial hyperscale data centre remains dramatically less expensive than launching an equivalent amount of hardware into orbit. Power infrastructure already exists. Maintenance crews remain readily available. Cooling technologies are mature, supply chains are established, and expansion requires little more than additional land and electrical capacity.</p><p>For many workloads, Earth will almost certainly remain the most practical location for computation.</p><p>Yet history repeatedly demonstrates that economics evolve alongside technology.</p><p>Only a generation ago, launching reusable rockets was widely considered unrealistic. Today, booster recovery has fundamentally altered the cost structure of access to space. Similar transitions may occur across spacecraft manufacturing, robotic servicing, autonomous operations, and in-space assembly.</p><p>More importantly, orbital computing may not need to compete directly with terrestrial infrastructure.</p><p>Instead, it may complement it.</p><p>Certain applications naturally benefit from operating beyond Earth’s atmosphere.</p><p>Large-scale processing of satellite imagery, climate observations, astronomical data, deep-space communications, and autonomous spacecraft operations all generate enormous quantities of information that currently must be transmitted to Earth before meaningful analysis can begin.</p><p>Processing this data directly in orbit could significantly reduce latency, lower communication demands, and enable entirely new classes of scientific and commercial missions.</p><p>As artificial intelligence becomes increasingly integrated into every aspect of space exploration, keeping computation closer to the source of the data becomes an increasingly compelling proposition.</p><p>The first orbital data centres may therefore emerge not because they replace terrestrial cloud computing, but because they solve problems that Earth-based infrastructure cannot address as efficiently.</p><h3>Final Thoughts</h3><p>Every major technological shift begins by challenging an assumption that once appeared permanent.</p><p>For centuries, manufacturing belonged beside rivers. Later, electricity liberated factories from geography. The internet removed many constraints on communication, while cloud computing redefined how software was built and deployed.</p><p>Orbital computing represents another possible inflexion point.</p><p>Its greatest obstacle is not a lack of ambition but the uncompromising realities of physics. Every subsystem must function within an environment that offers no atmosphere for cooling, exposes electronics to relentless radiation, limits available power, complicates communication, and renders routine maintenance almost impossible.</p><p>Yet none of these challenges appears fundamentally insurmountable.</p><p>Instead, they demand innovations across multiple disciplines- materials science, semiconductor engineering, thermal physics, robotics, communications, aerospace engineering, and artificial intelligence itself.</p><p>Whether orbital data centres become commonplace within the next decade or remain a longer-term ambition is still uncertain. What is becoming increasingly clear, however, is that the conversation has shifted. The question is no longer whether computation <em>can</em> occur in space- it already does. The real challenge is determining how to scale that capability into infrastructure capable of supporting the computational demands of an increasingly AI-driven world.</p><p>If that challenge is solved, future generations may view terrestrial data centres much as we now view rooms filled with vacuum tubes: essential stepping stones that ultimately gave way to something far more ambitious.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=74cb89c57950" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Why Manufacturing Should Start Thinking Like Silicon Valley]]></title>
            <link>https://medium.com/@kavyarathi1102/why-manufacturing-should-start-thinking-like-silicon-valley-f7f7753e5f6d?source=rss-e9b9ce957cf4------2</link>
            <guid isPermaLink="false">https://medium.com/p/f7f7753e5f6d</guid>
            <category><![CDATA[innovation]]></category>
            <category><![CDATA[silicon-valley]]></category>
            <category><![CDATA[business]]></category>
            <category><![CDATA[leadership]]></category>
            <category><![CDATA[manufacturing]]></category>
            <dc:creator><![CDATA[Kavya Rathi]]></dc:creator>
            <pubDate>Mon, 29 Jun 2026 16:35:16 GMT</pubDate>
            <atom:updated>2026-06-29T16:35:16.427Z</atom:updated>
            <content:encoded><![CDATA[<p>Walk into a factory and walk into a technology start-up, and you’ll notice the differences almost immediately.</p><p>One is built around precision, routine, and consistency. The other thrives on experimentation, uncertainty, and constant change.</p><p>For years, these industries have been treated as opposites. Manufacturing has often been viewed as traditional and methodical, while Silicon Valley has become synonymous with disruption and innovation. Yet as industries become increasingly connected through automation, artificial intelligence, and digital technologies, that distinction is beginning to fade.</p><p>The future of manufacturing will not be shaped solely by better machines. It will be shaped by better thinking.</p><p>The most valuable lesson Silicon Valley offers is not how to build software. It is how to build organisations that never stop improving.</p><p><strong>Innovation Begins Long Before the Product</strong></p><p>Many people assume innovation starts in a research laboratory or during product development. More often than not, it begins with a question.</p><p>Could this process be simpler?</p><p>Is there a better way of solving this problem?</p><p>What would happen if we started again from scratch?</p><p>Companies in Silicon Valley have built entire businesses around asking uncomfortable questions. They challenge assumptions that everyone else has accepted, even when those assumptions have worked for years.</p><p>Manufacturing has traditionally rewarded consistency, and for good reason. Customers expect reliability, safety, and quality. However, consistency should never become an excuse for resisting change.</p><p>The businesses that continue to grow are usually the ones willing to examine familiar processes with fresh eyes. Sometimes the greatest improvements come not from revolutionary inventions but from questioning routines that no one else has thought to challenge.</p><p><strong>Speed Has Become a Competitive Advantage</strong></p><p>In the past, size often determined success. Larger companies had greater resources, stronger supplier networks, and more established reputations.</p><p>Today, speed increasingly matters just as much.</p><p>Technology companies are known for releasing early versions of products, gathering feedback, and making improvements continuously rather than waiting for perfection. Their philosophy is simple: learn quickly, improve quickly.</p><p>Manufacturing cannot adopt this approach in exactly the same way. Physical products carry greater risks and higher production costs than software updates. Nevertheless, the principle still applies.</p><p>Whether testing a prototype, introducing a new production method, or trialling digital tools within a single facility, learning through carefully controlled experimentation allows companies to improve before committing significant resources.</p><p>Being cautious is sensible.</p><p>Being slow simply because “that’s how it’s always been done” is far more dangerous.</p><p><strong>The Best Decisions Rarely Come from Assumptions</strong></p><p>Every organisation collects information. The question is whether it actually uses it.</p><p>Modern factories generate enormous amounts of operational data, from machine performance and maintenance records to supply chain activity and product quality. Too often, however, these figures are reviewed only after problems have emerged.</p><p>Silicon Valley has demonstrated the value of making information part of everyday decision-making rather than something reserved for monthly reports or annual reviews.</p><p>When leaders understand what is happening in real time, they can identify trends sooner, solve problems earlier, and make decisions with greater confidence.</p><p>Experience will always matter.</p><p>But experience supported by evidence is considerably more powerful than experience alone.</p><p><strong>Technology Doesn’t Create Change-People Do</strong></p><p>Artificial intelligence, robotics, and automation dominate conversations about the future of manufacturing.</p><p>Yet technology itself changes very little unless people know how to use it effectively.</p><p>Some of the world’s most successful companies invest just as heavily in developing people as they do in developing products. They encourage collaboration between different disciplines, create opportunities for continuous learning, and allow employees to challenge existing ideas without fear of failure.</p><p>Manufacturing faces no shortage of complex problems worth solving. Improving sustainability, strengthening supply chains, reducing waste, and increasing efficiency are challenges that attract exceptionally talented engineers, scientists, and analysts.</p><p>Creating an environment where those people can genuinely influence decisions is just as important as hiring them in the first place.</p><p>Talent flourishes where trust exists.</p><p><strong>Adaptability Is Becoming a Core Business Skill</strong></p><p>Perhaps the greatest misconception surrounding successful companies is that they somehow predict the future more accurately than everyone else.</p><p>In reality, they usually adapt more quickly when circumstances change.</p><p>Markets shift. Customer expectations evolve. New competitors appear. Technologies that once seemed experimental become standard practice within surprisingly short periods of time.</p><p>Companies that remain flexible are far more likely to thrive because they are comfortable adjusting their strategies when new information emerges.</p><p>Adaptability should not be mistaken for abandoning long-term goals. Rather, it reflects a willingness to change the route without losing sight of the destination.</p><p>That quality has become increasingly valuable regardless of industry.</p><p><strong>Looking Beyond the Machines</strong></p><p>Manufacturing is often associated with equipment, production lines, and technical expertise.</p><p>Yet behind every successful factory is a series of human decisions.</p><p>Someone chooses where to invest.</p><p>Someone decides whether to embrace a new technology.</p><p>Someone determines how employees are supported, how risks are managed, and how improvement is measured.</p><p>These choices shape an organisation far more profoundly than any individual machine ever could.</p><p>Silicon Valley’s influence has never rested solely on technological breakthroughs. Its greatest contribution has been demonstrating what organisations can achieve when curiosity, learning, and adaptability become part of everyday culture.</p><p>Manufacturing does not need to resemble a technology start-up.</p><p>It does, however, have every reason to embrace the mindset that has allowed so many innovative companies to remain resilient in an increasingly unpredictable world.</p><p>Because in the end, competitive advantage is rarely defined by what a company makes.</p><p>More often, it is defined by how willing it is to learn.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=f7f7753e5f6d" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[The Psychology of Change: Why People Resist What They Need Most]]></title>
            <link>https://medium.com/@kavyarathi1102/the-psychology-of-change-why-people-resist-what-they-need-most-e1b6c6743a69?source=rss-e9b9ce957cf4------2</link>
            <guid isPermaLink="false">https://medium.com/p/e1b6c6743a69</guid>
            <category><![CDATA[productivity]]></category>
            <category><![CDATA[self-improvement]]></category>
            <category><![CDATA[leadership]]></category>
            <category><![CDATA[psychology]]></category>
            <dc:creator><![CDATA[Kavya Rathi]]></dc:creator>
            <pubDate>Wed, 27 May 2026 07:03:16 GMT</pubDate>
            <atom:updated>2026-05-27T07:03:16.910Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*AjJcaLjhZYAT4PAEIPGnmg.png" /></figure><p>George Bernard Shaw once wrote, <em>“Progress is impossible without change, and those who cannot change their minds cannot change anything.”</em> It is a statement that feels increasingly relevant in today’s world, where innovation moves faster than our willingness to adapt to it.</p><p>We live surrounded by change. Seasons transform, cities evolve, technologies replace one another almost overnight, and people themselves are constantly shaped by experience. Change is not an interruption to life — it is life. Yet despite understanding this intellectually, organisations and individuals still resist it with remarkable intensity.</p><p>What fascinates me is that this resistance rarely comes from irrationality or bad intentions. In fact, two observations seem consistently true across almost every environment:</p><ul><li>Change is inherent to nature.</li><li>People are inherently well-intentioned.</li></ul><p>At first glance, these ideas appear contradictory to the conflicts we witness in workplaces and institutions. If change is natural, why do people oppose it? And if people are fundamentally trying to do good, why do organisations experience so much internal friction?</p><p>The answer lies not in human incompetence, but in human psychology.</p><h3>Why Resistance Exists</h3><p>In nature, transformation occurs quietly and continuously. A sapling matures into a tree; autumn leaves decay to nourish the next season’s growth. There is no resistance because nature does not fear uncertainty.</p><p>Humans, however, do.</p><p>Within organisations, people build routines around familiarity. Over time, these routines become associated with competence, stability, and identity. A system that feels predictable becomes psychologically safe. Any disruption to that system — even one designed to improve it — can feel threatening.</p><p>What appears externally as resistance is often internally a fear of losing certainty.</p><p>This fear is usually reinforced by experience. Failed initiatives, poorly managed transitions, or previous instability condition people to associate change with risk rather than opportunity. Consequently, organisations develop cultures where preserving comfort becomes more important than pursuing progress.</p><p>This explains why many change initiatives fail before implementation has even begun.</p><h3>Why Conflict Exists Even Among Good People</h3><p>Eliyahu M. Goldratt, creator of the Theory of Constraints, argued that people are inherently good — meaning most individuals genuinely believe they are contributing positively to the system around them.</p><p>I believe this is largely true.</p><p>Very few people arrive at work intending to create inefficiency or failure. Most decisions, even flawed ones, are made with the belief that they are beneficial.</p><p>Yet conflict persists because individuals interpret problems through the narrow lens of their own responsibilities.</p><p>In a hospital, for example, a surgeon may believe the core issue is understaffing. A nurse may attribute problems to insufficient equipment. An IT specialist may view outdated software as the primary obstacle. None of these perspectives is necessarily wrong, but each represents only one fragment of a larger system.</p><p>Organisations fail when people become more committed to defending perspectives than understanding causes.</p><p>Blame becomes easier than analysis.</p><p>However, meaningful change only becomes possible when leaders move beyond assigning fault and instead challenge the assumptions driving decisions in the first place.</p><h3>The Failure of Change Initiatives</h3><p>Most change programmes fail not because the solution itself is weak, but because the process of introducing it is incomplete.</p><p>Selling change is not fundamentally different from selling an idea, a vision, or even a product. It requires sequence, trust, and psychological alignment.</p><p>Too often, organisations attempt to introduce solutions before people have emotionally accepted the existence of a problem.</p><p>According to Goldratt’s framework, successful transformation begins with four essential questions.</p><h4>1. Why Must We Change?</h4><p>This is the foundation of every successful transformation.</p><p>Before people support a solution, they must first believe that remaining the same is no longer sustainable. If stakeholders feel comfortable with the current state, even the most sophisticated proposal will feel unnecessary.</p><p>People change only when they perceive a meaningful gap between present reality and future potential.</p><p>Without urgency, there is no momentum.</p><h4>2. What Exactly Needs to Change?</h4><p>This stage is significantly more complex than it appears.</p><p>Different groups within an organisation often define the problem differently because each experiences only a portion of the system. Sales may blame operations; operations may blame management; management may blame market conditions.</p><p>The challenge is therefore not merely identifying symptoms, but uncovering the underlying issue connecting them all.</p><p>This requires deep listening, systemic thinking, and the willingness to question assumptions rather than individuals.</p><p>Consensus on the problem must exist before consensus on the solution can emerge.</p><h4>3. What Should We Change To?</h4><p>Even when people agree that a problem exists, resistance often returns once solutions are proposed.</p><p>Why?</p><p>Because uncertainty intensifies when consequences become tangible.</p><p>At this stage, people begin asking:</p><ul><li><em>Will this actually work?</em></li><li><em>What if the solution creates new problems?</em></li><li><em>Is this truly the best path forward?</em></li></ul><p>These questions are not signs of negativity; they are signs of investment.</p><p>Strong leaders understand that persuasion is not achieved through authority alone. People support change when they understand both the logic behind it and the value it creates for them and the wider system.</p><p>This stage demands patience, transparency, and intellectual honesty.</p><h4>4. How Do We Make the Change Happen?</h4><p>Only after the previous stages are resolved should implementation begin.</p><p>This is where practical and emotional barriers emerge simultaneously.</p><p>People may agree with the concept while still fearing the consequences of execution:</p><ul><li>disruption,</li><li>instability,</li><li>resource limitations,</li><li>reputational risk,</li><li>or failure itself.</li></ul><p>This is the classic <em>“Yes, but…”</em> phase of change.</p><p>A proposal may appear excellent theoretically while still feeling impossible operationally.</p><p>Effective change leaders do not dismiss these concerns. They address them systematically, refine plans where necessary, and create environments where people feel secure enough to move forward despite uncertainty.</p><p>Successful execution depends not only on strategic planning but on emotional credibility.</p><h3>Conclusion</h3><p>The paradox of change is that humanity simultaneously depends upon it and resists it.</p><p>We admire innovation, yet instinctively protect familiarity. We demand progress, yet hesitate when progress requires discomfort.</p><p>The art of selling change, therefore, is not about forcing compliance. It is about creating understanding — helping people see not only why transformation is necessary, but why remaining unchanged may be the greater risk.</p><p>In an era defined by constant reinvention, adaptability has become one of the most valuable forms of intelligence. Organizations, institutions, and individuals who learn to evolve thoughtfully will continue to grow. Those who refuse eventually become irrelevant.</p><p>Change is inevitable.</p><p>Convincing people to believe in it — that is the true challenge.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=e1b6c6743a69" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>