<?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 Sreeram Narasimhan on Medium]]></title>
        <description><![CDATA[Stories by Sreeram Narasimhan on Medium]]></description>
        <link>https://medium.com/@mareers?source=rss-d82d3ee616ff------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*omPaJVAIGPMAGbN75dolnA.jpeg</url>
            <title>Stories by Sreeram Narasimhan on Medium</title>
            <link>https://medium.com/@mareers?source=rss-d82d3ee616ff------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Thu, 19 Nov 2026 23:10:58 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@mareers/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[You’re Already Paying for the AI That Knows Your Entire Work History]]></title>
            <link>https://medium.com/naturally-stupid/youre-already-paying-for-the-ai-that-knows-your-entire-work-history-d222a5b1fa21?source=rss-d82d3ee616ff------2</link>
            <guid isPermaLink="false">https://medium.com/p/d222a5b1fa21</guid>
            <category><![CDATA[leadership]]></category>
            <category><![CDATA[productivity]]></category>
            <category><![CDATA[future-of-work]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[microsoft]]></category>
            <dc:creator><![CDATA[Sreeram Narasimhan]]></dc:creator>
            <pubDate>Fri, 24 Apr 2026 11:17:50 GMT</pubDate>
            <atom:updated>2026-04-24T11:18:48.833Z</atom:updated>
            <content:encoded><![CDATA[<p>Microsoft Copilot has been getting a lot of flak on social media recently, and honestly, not without reason. Anyone who has used Claude inside Excel or PowerPoint and then gone back to Copilot on the same tasks will feel the gap. I have been on both sides of this, and the criticism I see online matches what I have experienced personally.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*JzY2W7i5gEMHFdBn3dzJEA.png" /></figure><p>But here is what that conversation keeps missing.</p><h3><strong>The scale nobody is walking away from</strong></h3><p>Microsoft 365 has 450 million paid commercial seats globally, growing 6% year-on-year, with another 90 million consumer subscribers on top. GitHub Copilot crossed 4.7 million paid subscribers as of January 2026, inside 90% of the Fortune 100. Microsoft 365 Copilot hit 15 million paid seats by December 2025, up 160% year-on-year, with reportedly 90% of the Fortune 500 using it in some form.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*bNEt1dFo01uCv5IWsP4NTg.png" /></figure><p><em>(Figures from Microsoft’s FY26 Q2 earnings call, January 28, 2026. Q3 figures covering January–March 2026 are due April 29, 2026.)</em></p><p>The honest number in all of that is 3.3%: the share of those commercial seats that have actually converted to paid Copilot after two years of availability. Enterprises are not rushing to move away from the Microsoft stack, but they are not rushing toward the AI layer either. The procurement cycles are long, the integrations are deep, and the switching costs are real. Copilot does not have to be the best model in the room. It just has to be in the room.</p><p>Microsoft has been here before. When Teams launched in 2017, it was widely considered the worse product. Slack had better UX, stronger developer tooling, and a genuinely enthusiastic user base. Microsoft had Office 365 and hundreds of millions of seats to drop Teams into for free. That was enough. Slack lost the enterprise, not because it lost the product fight, but because it lost the distribution fight. Microsoft will be looking to take a leaf out of that playbook for the Copilot journey.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*u665e7pLvdis8rMb6QnhRg.jpeg" /><figcaption><strong>Note: </strong>Teams figures after 2021 are Monthly Active Users (MAU). Slack remains DAU throughout. <strong>Data: </strong>Company reports &amp; Statista (as of April 2026).</figcaption></figure><p>One data point from this week: GitHub paused new sign-ups for Copilot Pro and Pro+ on April 20, citing unsustainable compute costs from agentic workloads. The installed base story does not change, but the economics of enterprise AI are still being worked out in real time.</p><h3><strong>The unfair advantage: Microsoft Graph</strong></h3><p>There is one area where Microsoft has something no other AI model can replicate without an MCP connector or a custom integration. (Claude <a href="https://x.com/claudeai/status/2040086268562842097?s=20">recently announced</a> Microsoft connectors for Outlook and files, which tells you something about where the gap is and who is working to close it.) It is called Microsoft Graph, and it is the reason I think Copilot still wins on a specific class of problems.</p><p>Microsoft Graph is essentially a unified, real-time record of everything you do inside Microsoft 365. Every email you send in Outlook, every meeting on your Teams calendar, every document you open in SharePoint, every chat message in Teams, every task in Planner: all of it indexed in one place, linked to your identity, timestamped.</p><p>More specifically, it holds:</p><ul><li><strong>Your email:</strong> who you communicate with, how often, on what topics</li><li><strong>Your calendar:</strong> who you meet, recurring patterns, time commitments</li><li><strong>Your files:</strong> what you create, edit, and share, and with whom</li><li><strong>Teams activity:</strong> chats, calls, meeting notes, channel participation</li><li><strong>Your people graph:</strong> your closest collaborators, reporting line, org position</li><li><strong>Viva Insights signals:</strong> focus time, after-hours load, collaboration patterns</li></ul><p>When you ask Copilot “What did Priya say she would do before the deadline?”, it is not searching the internet. It is querying your personal graph: pulling the Teams transcript from the meeting on your calendar, cross-referencing the attendees, surfacing the relevant email thread from the same week. No other AI has this level of contextualised access to your work life without additional setup.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/800/0*CMxYldzL6d8C4TdH.png" /><figcaption><strong>Source:</strong> <a href="https://learn.microsoft.com/en-us/graph/overview">Overview of Microsoft Graph</a></figcaption></figure><p>One important note for the privacy-conscious: Microsoft Graph does not expose your data to other users, or to Microsoft’s AI training. It operates within your organisation’s existing security boundary. Copilot can only surface data you already have permission to see.</p><p>There is a powerful layer already live on top of all of this. Microsoft has rolled out <strong>Communication Memory</strong>, an AI system that continuously analyses and summarises your emails, Teams chats, and meeting transcripts into a running, cross-channel context stored privately in your mailbox. Once enabled in your tenant, asking Copilot a question in Word lets it pull a thread from an Outlook conversation that happened weeks ago, without you naming the source or knowing where to look. The feature became generally available in September 2025 and is progressively strengthening Copilot’s ability to deliver deeply personalised, graph-grounded answers across Microsoft 365.</p><h3><strong>What this actually means for how you work</strong></h3><p>I have been using Copilot for a set of use cases that are only possible because of this graph access. No connectors, no prompt engineering, no custom setup. These work out of the box.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*-O0ISOow6Mjl367HTnw9tA.png" /></figure><h4><strong>Morning clarity</strong></h4><p>Ask Copilot what is on your plate today: meetings, pending emails that need a response, tasks that are overdue. It pulls from your calendar, your inbox, and your Planner simultaneously. Five minutes instead of twenty.</p><h4><strong>End of day accounting</strong></h4><p>Ask Copilot what you actually accomplished today. It can piece together the meetings you attended, the documents you worked on, and the emails you sent. Useful for tracking, and surprisingly accurate as a reflection of where time actually went versus where you thought it went.</p><h4><strong>Meeting preparation</strong></h4><p>Before a client call or an internal review, ask Copilot to brief you on the attendees: recent emails exchanged, what was discussed in your last meeting, any open action items. It pulls this from your graph without you having to dig through Outlook or search meeting notes. Particularly useful in presales and account management, where the preparation matters as much as the meeting itself.</p><h4><strong>Understanding your collaboration patterns</strong></h4><p>Ask who you have been spending the most time with over the last month. Copilot can map this from your calendar and Teams activity. For people managers especially, this is a useful check on whether your attention is going where you intend it to.</p><h4><strong>Relationship summary for accounts</strong></h4><p>For anyone in sales or presales: ask Copilot to summarise your interactions with a specific company over the last six months. It draws on emails, meeting notes, and shared documents within your account, within existing permissions. No CRM query needed. No chasing colleagues for context. The history is already there.</p><h4><strong>Appraisal documents, and something bigger than that</strong></h4><p>This is the use case that actually prompted this post, but I want to frame it properly because the appraisal form is really just the symptom of a deeper problem.</p><p>Most knowledge workers will recognise this feeling: the day runs away without seemingly much being achieved. It is not that you were idle. You were in meetings, responding to emails, reviewing documents, unblocking things for other people. But at the end of the day, you cannot point to what you accomplished. And this feeling does not stay contained to a single day. Days turn into weeks, weeks turn into months, and the quiet sense of being unproductive just keeps accumulating in the background.</p><p>Then appraisal season arrives and you are suddenly expected to account for twelve months of work. Most people end up scrambling: searching old email threads, trying to remember which presentations they built, piecing together a case for their own contribution. It is an exhausting process that relies entirely on memory, which is exactly the wrong tool for the job.</p><p>This is where Copilot becomes genuinely useful. Not as an appraisal rescue tool, but as a daily habit. Ask it what you accomplished today. Ask it what is ahead tomorrow. Ask it who you have been collaborating with this quarter. Build this into your routine and by the time appraisal comes around, you are not reconstructing your year from scratch. You already have a running picture of it.</p><p>Here is a twist I did not expect when I started using it this way: I do not actually use Copilot to fill up my own appraisal form. I use it to understand what my reportees have been working on. Before their review conversations, I ask Copilot to surface the projects they have contributed to, the people they have collaborated with, the volume and patterns of their work over the quarter. It gives me a much fuller picture than what makes it into a self-appraisal document, and it levels the playing field for the people who are not great at writing about themselves.</p><p>The bigger reimagining here is this: Copilot as a daily productivity driver, not a once-a-year document filler. Morning focus, evening reflection, weekly patterns, quarterly view. All without a separate tool, a new subscription, or an MCP integration. It is already inside the Microsoft 365 you are using every day.</p><h3><strong>The point is not that Copilot is the best AI</strong></h3><p>It is not, at least not on the kind of general reasoning and creative tasks where Claude or GPT-4o pull ahead. That race is ongoing and Microsoft knows it.</p><p>The point is that Microsoft holds a comprehensive, personalised, timestamped picture of your entire working life inside their ecosystem. And most users are not actively using this to their advantage. The tools to do so are already in your Microsoft 365 subscription.</p><p>The context is already there. The question is whether you use it.</p><h3><strong>Key Takeaways</strong></h3><ul><li>Copilot’s edge is not a better model. It is the depth of what Microsoft Graph already knows about your working life.</li><li>The real value comes from building daily habits: morning briefing, evening reflection, not from pulling it out once a year at appraisal time.</li><li>The data is already sitting in your Microsoft 365 tenant. Most people just never ask for it.</li></ul><h3><strong>Appendix: Prompts and practical notes</strong></h3><p>The use cases in this post work today with the paid Microsoft 365 Copilot license ($30/user/month enterprise, $21 SMB). The free Copilot Chat handles web-grounded queries but cannot read your mailbox, calendar, or Teams history. Toggle to Work mode (the green shield) at m365.cloud.microsoft/chat before using any of these.</p><h4><strong>Morning briefing</strong></h4><p>The Viva Insights Briefing email, which many readers will remember as the “here’s your day” digest, was paused in December 2022 and has never come back. The replacement is on-demand BizChat or a Scheduled Prompt set to run at 8 a.m. daily: hover over any response, click the clock icon, set the cadence.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*NkjMddYzj8rwXvQEpquV2w.png" /></figure><h4><strong>End of day</strong></h4><p>The same BizChat surface, used at 5 p.m. Narrow prompts with explicit date windows produce much more accurate results than broad ones. A Scheduled Prompt set to 17:00 on weekdays automates this entirely.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*SUsj81-roxoXj9uoRVt-uA.png" /></figure><h4><strong>People intelligence</strong></h4><p>For numbers, use the Viva Insights app (Teams or insights.cloud.microsoft) rather than BizChat. The Network page shows a three-ring collaboration map over the last four weeks, with time spent broken down by email, meetings, chats, and calls. This does not require a Copilot licence, only an Exchange Online-enabled subscription.</p><p>For conversational queries, BizChat in Work mode will answer, but treat the results as approximations and verify anything consequential in Viva Insights.</p><ul><li>“Who have I been collaborating with most this month?”</li><li>“Summarise my interactions with [person] in the past 30 days across emails, chats, and meetings.”</li><li>“What did I discuss with the [team] in April?”</li></ul><h4><strong>Appraisal season</strong></h4><p>The workflow that holds up best is four steps: gather evidence in BizChat, synthesise in a Copilot Notebook, draft in Word, polish with Rewrite.</p><p>Gathering evidence (use narrow prompts with explicit dates, not one omnibus prompt):</p><ul><li>“Summarise my key accomplishments from January through March 2026 based on my emails, Teams messages, meetings I led, and files I authored.”</li><li>“Find emails from my manager or stakeholders that contain praise, thanks, or positive feedback since October 2025.”</li><li>“What meetings did I organise or lead in the past quarter? Group by project.”</li><li>“Based on my calendar and Teams activity, estimate how I spent my time across my top initiatives.”</li></ul><p><strong>Drafting in Word:</strong> reference the Notebook or attach up to three files with /. A prompt that works well: &quot;Draft a performance review self-assessment for FY2026 based on [notebook]. Use STAR format. Around 600 words. Professional, confident but humble tone. Include measurable outcomes where possible.&quot;</p><p><strong>One important limitation to know: </strong>Copilot’s retrieval biases strongly toward the last 90 days of your Graph activity. The fix is a monthly habit. On the last Friday of each month, run your accomplishment prompts, click “Edit in Page,” and save a Copilot Page titled “Accomplishments: [Month] 2026” to a dedicated SharePoint folder. At review time, reference all 12 pages in one Notebook. This is the single highest-leverage thing you can do if you plan to use Copilot at appraisal time.</p><h4><strong>Free vs paid</strong></h4><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*SblVZto5V4E355xF4zd3pA.png" /></figure><h4><strong>Things that no longer exist</strong></h4><p>Two features cited in older articles are no longer available. Viva Goals, the OKR tool, was retired on December 31, 2025 with no Microsoft replacement. If your team tracked objectives there, you need a third-party tool or a manual SharePoint/OneNote setup. The Viva Insights Briefing email has been paused since December 2022 and has not returned; the Scheduled Prompts approach above is the practical substitute.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=d222a5b1fa21" width="1" height="1" alt=""><hr><p><a href="https://medium.com/naturally-stupid/youre-already-paying-for-the-ai-that-knows-your-entire-work-history-d222a5b1fa21">You’re Already Paying for the AI That Knows Your Entire Work History</a> was originally published in <a href="https://medium.com/naturally-stupid">Naturally Stupid</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[The Intelligence Decade: What the 6G-AI Convergence Actually Means for Business Leaders]]></title>
            <link>https://medium.com/naturally-stupid/the-intelligence-decade-what-the-6g-ai-convergence-actually-means-for-business-leaders-f57980e195c0?source=rss-d82d3ee616ff------2</link>
            <guid isPermaLink="false">https://medium.com/p/f57980e195c0</guid>
            <category><![CDATA[technology]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[telecommunication]]></category>
            <category><![CDATA[6g]]></category>
            <category><![CDATA[digital-transformation]]></category>
            <dc:creator><![CDATA[Sreeram Narasimhan]]></dc:creator>
            <pubDate>Mon, 20 Apr 2026 05:08:00 GMT</pubDate>
            <atom:updated>2026-04-20T05:08:23.698Z</atom:updated>
            <content:encoded><![CDATA[<p>I remember having to dial up the internet to check my school board exam results in 1999 (and once again in 2001, if memory serves). The modem screech. The agonising wait while the page loaded, line by painful line. You “went online” — deliberately, conscious of the act. When you were done, you came back offline.</p><p>Today those results are integrated with a student’s Aadhaar ID and accessible via DigiLocker. We do not think about internet connectivity when looking up exam results. At worst, we worry about web traffic slowing the website during peak season. The infrastructure became invisible. Like electricity, you notice it only when it is gone.</p><p>BCG published four interconnected reports between January and March 2026 covering AI enterprise adoption, telco transformation, 6G infrastructure, and the technology services market. Here is my honest read of what they mean. I am happy to be proven wrong — prediction is not an exact skill.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*vy1XzHtUuujD5UdPsE0h8A.png" /></figure><h3>1. We stopped “going online.” The internet went inside us.</h3><p>The Aadhaar and DigiLocker examples above are not just convenient illustrations. They represent a structural shift in how we relate to connectivity. In 2024, US data consumption hit 132 trillion MB. Mobile broadband now covers 96% of the global population. The numbers are large, but the real signal is cultural: we stopped noticing connectivity.</p><p>The 6G era takes this further. The network will not just carry your data. It will process it, sense the environment around you, and act on your behalf. The question for business leaders is not whether this transition is coming. It is which parts of your operating model still assume a world where connectivity is a variable rather than a given.</p><h3>2. The uplink revolution nobody is talking about</h3><p>The internet was built for consumption. Thirty years of network design rested on a single assumption: data flows downward, from servers to screens, from studios to audiences. Traditional networks were built 90% for download.</p><p>AI is reversing this — and the data is already showing up on live networks. Ericsson’s 2025 Mobility Report measured AI application traffic on operating networks and found the ratio has already shifted to 74% downlink, 26% uplink. AI glasses already generate a 1:8 upload-to-download ratio, uploading eight times more data than they receive. Nokia has warned of a potential uplink surge that could overwhelm current infrastructure if operators do not plan for it now.</p><p>But there is a more human version of this shift playing out alongside the infrastructure numbers. YouTube receives over 700 hours of video every minute. TikTok sees roughly 16,000 videos uploaded per minute. Canva grew from around 40 million users in 2020 to approximately 220 million by 2025, more than five times in five years, largely on the back of non-designers creating content they previously could not make. A lot of people are turning from content consumers to content creators. X showed how social media upended traditional media in terms of bringing the most breaking news first. A phone or laptop, always connected, with agentic powers to create, edit, publish, and distribute, will potentially turn everyone into a producer. Not just in media. In every domain.</p><p>Current 5G networks were not designed for a world where every device broadcasts continuously. 6G is the redesign for that world. The business implication is equally real: if your operating model assumes customers are passive recipients of your content and services, that assumption is already ageing.</p><h3>3. That $18.2 trillion number needs context</h3><p>BCG and Qualcomm project $18.2 trillion in cumulative economic value from 5G/6G convergence by 2035. As a directional indicator of the scale of the platform shift, it does that job. But before anchoring any strategy to it, it is worth checking the previous forecast.</p><p>IDC’s 2021 IoT market forecast predicted $8.3 billion by 2026. Actual 2026 revenue landed closer to $5.88 to $7.57 billion. Still a massive CAGR of 35 to 49%. Still a substantial market. But lower than the headline predicted, and delivered across a longer runway than the original model assumed.</p><p>Large technology platform projections consistently outrun near-term adoption while underestimating the eventual long-term effect. We are at such an early stage of a technology concept that any number can end up being a finger in the air. The technology could turn in a different way when theories meet practical implementations. The $18.2 trillion is a signal of order of magnitude, not a planning input.</p><blockquote>The more useful question is not “how do I get a share of $18.2 trillion” but “what is the first concrete slice of value I can create and capture, right now, with the infrastructure that actually exists?”</blockquote><h3>4. Why most AI projects are failing at the wrong layer</h3><p>BCG’s most practically useful finding is what they call the 10–20–70 rule: 10% of effort on algorithms, 20% on technology and data, 70% on people, process, and change management. Most organisations are stuck in the 10% conversation, debating which model to use rather than asking whether the organisation can actually change how it works.</p><p>I have seen this play out, not in AI specifically, but in IoT and sustainability work over 2020–22. In that period, Azure and AWS invested heavily in IoT product suites, with partnerships, certifications, and go-to-market activity ramping up significantly on both sides. The technology investment was real. But by 2023 and 2024, both platforms began pulling back. Google Cloud IoT Core was shut down in August 2023. Microsoft announced and then partially walked back a retirement notice for Azure IoT Central in early 2024, signalling that neither hyperscaler had found a sustainable path in cloud IoT platforms. The enterprise OT sales cycle simply does not match a cloud infrastructure business model. And the organisational change required on the customer side proved harder to engineer than anyone had planned.</p><p>The private 5G story followed a similar arc. In 2021, the narrative was that it would instantly replace all industrial Wi-Fi and cabling. By Q3 2025, there were over 3,700 documented private 5G installations globally, which is meaningful progress. But the model that emerged was coexistence: 5G for robotics, motion control, high-resolution video, and mobile assets; Wi-Fi for laptops and low-criticality devices; Ethernet for ultra-critical systems. Not replacement. Segmentation. An NTT Data survey from 2021 found 90% of executives expected private 5G to become the new network standard. Actual deployment was, in the industry’s own words, “far more sluggish than anticipated.” The shift to commercial-scale production deployments only began in earnest in 2024–25.</p><p>There are success stories. CJ Logistics in South Korea replaced 300 Wi-Fi access points with private 5G at a 40,000 sqm warehouse and reported a 20% productivity increase. Delta Electronics in Taiwan deployed private 5G on smart production lines and reported significant output per headcount gains. The technology works when the conditions are right. The problem is that most organisations focus on getting the technology right and underestimate everything else that needs to align simultaneously.</p><p>Looking across these examples and earlier platform shifts, a pattern emerges. The transformations that worked were not just technologically ready. They had a forcing mechanism: a regulator mandate, a market disruption, or an external shock. They had organisations that genuinely rewired incentives and workflows, not just ran a pilot. They had an economic case that was felt immediately, not projected over five years. And they had a supporting ecosystem — standards, skills, supply chain — that was actually complete. When one or more of those conditions are missing, the technology stalls in the organisation regardless of how sound the underlying idea is.</p><p>The lesson is not that the technology was wrong. A lot of these ideas did take shape. The disruption felt more like organic growth from what was already there than the overnight transformation the original narrative promised. Implementation success hinges not only on a breakthrough technology but more importantly on getting the process and people aligned.</p><p>The clearest Indian example of all the conditions aligning is UPI. Adoption blew up not because the technology was revolutionary in isolation. It was the alignment of multiple factors: widespread mobile ownership, Reliance Jio crashing data prices, and Covid making social distancing a priority and cash transactions risky. That kind of aligning of stars is what converts a breakthrough technology into an everyday usage item. The BCG framework is correct. Realising it is harder in practice — particularly at a large organisation with legacy structures and decade-long capex cycles. 5G Americas estimates that core operations transformation in manufacturing and utilities takes five to seven years, independent of whether the network is technically live.</p><h3>5. The Superbrain telco: destination or aspiration?</h3><p>The cautionary tale is AT&amp;T. In 2018, they paid $85 billion for Time Warner, betting that vertical integration of content and connectivity would let them compete with Netflix and Google. By 2022, they had spun out WarnerMedia Discovery for roughly $43 billion, about half what they paid. The internet had already stripped telcos of voice and SMS revenue. Over-the-top apps commoditised their most valuable services. Moving into content was meant to be the answer. It wasn’t.</p><p>Some telcos are making genuine progress in a different direction. SK Telecom restructured their entire company around AI, disclosed AI revenue as a separate P&amp;L line, and their AI Data Centre business grew 34.9% year-on-year in 2025. They are the clearest proof-point that a different outcome is possible. But the IBM Institute for Business Value found that 77% of telecom executives believe AI is improving their responsiveness, while 65% admit their AI initiatives have not delivered the expected value. SK Telecom is the exception, not the rule. The industry’s best current attempt at monetising network intelligence, the GSMA’s Open Gateway initiative, had 73 operator groups representing 80% of the global mobile market signed up by H1 2025 — but GSMA itself acknowledged in H2 2024 that “the focus has now shifted to monetisation,” which tells you it had not happened at scale yet.</p><p>There is a possibility of telcos having a better say this time, if they are truly able to bake intelligence into the network and find a way to monetise the data. But that requires a strong disruptor willing to break existing business models, not just optimise them. That may or may not happen. In the absence of that, the most likely near-term beneficiaries are not the telcos themselves but the technology services firms helping them navigate the transformation, and the enterprises that figure out how to use the network as infrastructure for their own AI ambitions.</p><h3>One honest caveat to all of the above</h3><p>The pattern from IoT, from 5G, from every previous platform shift is consistent: the technology is usually right, the timeline is usually optimistic, and the organisations that extract real value are the ones that invest in the 70% nobody wants to talk about. <strong>People, process, change.</strong></p><p>But this is my read, formed from a specific vantage point at a specific moment. I have been on the cautious side of technology timelines before, and the tide has still turned — just slower than the opening narrative suggested. The private 5G coexistence story I described is real. But so are the 3,700 installations globally and the productivity gains at the companies that got all the conditions right.</p><p>The $18.2 trillion may end up being roughly correct. The Superbrain telco may arrive ahead of schedule. Prediction is not an exact skill. What I am more confident about is the process: the organisations that will benefit most from the Intelligence Decade are the ones building the 70% now, before the technology forces the question.</p><h3>For the curious: a brief concept reference</h3><p><em>Short explainers on the terms that appear throughout this post.</em></p><p><strong>Agentic AI:</strong> AI systems that do not just answer questions. They autonomously perceive their environment, reason over complex goals, and take action without step-by-step human direction. The key distinction from generative AI: generative AI is a creative engine that produces outputs; agentic AI is a strategic executor that pursues outcomes. The 30% of AI budget going to agentic tools in BCG’s survey is the most important number in their research.</p><p><strong>The 10-20-70 Rule:</strong> BCG’s framework for AI implementation: 10% of effort on algorithms (now largely commoditised), 20% on technology and data, 70% on people, process, and change management. MIT Sloan research found that in real agentic deployments, 80% of the actual work was consumed by data engineering, stakeholder alignment, governance, and workflow integration — not the AI model itself.</p><p><strong>Pilot Purgatory:</strong> The state most organisations find themselves in: running isolated AI proofs-of-concept that never scale to production. Deloitte’s 2025 research found 30% of organisations are still only exploring agentic options and 38% are merely piloting. The cause is almost always the neglected 70%.</p><p><strong>Uplink vs. Downlink:</strong> Traditional networks were optimised for download (downlink). Agentic AI and AR devices reverse this: they broadcast continuous sensor streams and context data upward (uplink). BCG and Qualcomm’s research shows this ratio has already shifted from 90/10 to 74/26, and it will intensify as agentic adoption scales. The network was not designed for what we are starting to ask of it.</p><p><strong>Network-as-a-Service (NaaS):</strong> A model where enterprises access connectivity on demand via APIs and pay for outcomes rather than owning infrastructure. The telco monetisation question is really a question of whether operators can build a credible NaaS layer before hyperscalers do it for them.</p><p><strong>The $200 Billion Tech Services Opportunity:</strong> BCG’s projection that agentic AI will expand the technology services market by $200 billion in five years. The new demand is for systems integration, workflow engineering, and knowledge codification — translating institutional human knowledge into structured data that an AI agent can act upon. This is one of the highest-value, hardest-to-automate services in the agentic era.</p><h3>Sources</h3><h4>Primary research</h4><ol><li>BCG / Qualcomm — 6G: The Network for the Future of AI (February 2026). Core source for 6G architecture, uplink/downlink ratio data, $18.2T projection, and ISAC. <a href="https://web-assets.bcg.com/9c/73/4ac7634848a0b0bb31cf8dae40cb/6g-the-network-for-the-future-of-ai-and-immersive-connectivity-report-feb-2026-edit-03-web.pdf">Full report (PDF)</a></li><li>BCG — AI Radar 2026 / As AI Investments Surge, CEOs Take the Lead (2026). CEO AI archetypes (Followers, Pragmatists, Trailblazers), enterprise AI spend doubling to 1.7% of revenues, agentic AI budget share. <a href="https://www.bcg.com/publications/2026/as-ai-investments-surge-ceos-take-the-lead">bcg.com</a></li><li>BCG — The AI-First Telecommunications Strategy (2026). 10–20–70 rule, Superbrain telco model, EBITDA uplift projections (8–15pp), Deploy/Reshape/Invent framework. <a href="https://www.bcg.com/publications/2026/the-ai-first-telecommunications-strategy">bcg.com</a></li><li>BCG — The $200 Billion AI Opportunity in Tech Services (2026). Agentic AI market expansion, knowledge codification, systems integration demand. <a href="https://www.bcg.com/publications/2026/the-200-billion-dollar-ai-opportunity-in-tech-services">bcg.com</a></li></ol><h4>Network and infrastructure data</h4><ol><li>Ericsson Mobility Report (June 2025). Live network measurement of AI traffic: 74% downlink / 26% uplink ratio; ChatGPT traffic profile; GenAI as 0.06% of total network traffic; 115 million AI app downloads in December 2024.</li><li>Ericsson — How AI Devices Are Driving Uplink-First Networks (March 2026). AI glasses generating 1:8 upload-to-download ratio; autonomous vehicles and drones following similar patterns; network architecture shift from downlink-dominant to uplink-intensive.</li><li>Nokia — Global Network Traffic 2030 Report. Uplink surge warning; AI traffic forecast of 1,088 exabytes per month by 2033; hybrid AI requiring more than 5–15 Mbps uplink available on current networks.</li><li>GSMA — Open Gateway H1 2025 and H2 2024 Reports. 73 operator groups (representing ~79% of global mobile subscribers) signed up; SIM Swap and Number Verify as leading commercial API deployments; monetisation stage acknowledgement in H2 2024.</li><li>5G Americas — AI at the Edge (June 2025). Core operations transformation in manufacturing and utilities estimated at five to seven years; GSMA projection of AI adding 20–80% additional mobile network traffic.</li><li>SNS Telecom &amp; IT — Private 5G Market Report (August 2025). 3,700+ documented global private 5G installations in Q3 2025; manufacturing as leading sector; CJ Logistics (South Korea) and Delta Electronics (Taiwan/Thailand) case studies.</li><li>NTT Data / NETSCOUT — Private 5G CIO Survey (circa 2022/2024). 90% of CIOs believed private 5G would become the new network standard; actual deployment “far more sluggish than anticipated.”</li></ol><h4>Platform and creator economy</h4><ol><li>Teleprompter.com — YouTube Statistics (2025). 500 hours of video uploaded per minute; 800 million total videos; $36.1 billion in ad revenue in 2024.</li><li>Buffer / BrowserAct — TikTok Statistics (2025). 16,000 videos uploaded per minute; $4.1 billion earned by creators in 2024.</li><li>Sacra / Backlinko / Demandsage — Canva Statistics (2025–2026). 260 million monthly active users (up from 30 million in 2020); 38.5 million designs created per day; 800 million AI tool uses per month.</li></ol><h4>Telco and technology industry</h4><ol><li>IBM Institute for Business Value — Telecom AI Research (November 2025). 77% of telecom executives report AI improving responsiveness; 65% admit AI initiatives have not delivered expected value.</li><li>SK Telecom — FY2025 Earnings Release (February 2026). AI Data Centre revenue: KRW 519.9 billion (up 34.9% YoY); AIX B2B segment; restructuring into seven divisions; AI Pyramid 2.0 strategy; March 2025 cybersecurity breach and KRW 134.8 billion fine.</li><li>Variety / Fortune — AT&amp;T / Time Warner (April 2022 / May 2021). $85 billion acquisition in 2018; WarnerMedia Discovery spinout at ~$43 billion in 2022; analysis of root causes.</li><li>The Register — Google Cloud IoT Core Shutdown (August 2022). Shutdown announced and executed August 2023; analyst commentary on product stagnation and enterprise OT sales mismatch.</li><li>The Register — Microsoft Azure IoT Central Retirement (February 2024). Retirement announced effective March 2027; no direct successor announced.</li></ol><h4>Frameworks and academic sources</h4><ol><li>MIT Sloan Management Review — Agentic AI Explained. Distinction between generative AI (output) and agentic AI (outcome); autonomous perception and action. <a href="https://mitsloan.mit.edu/ideas-made-to-matter/agentic-ai-explained">mitsloan.mit.edu</a></li><li>Deloitte Tech Trends 2026 — Agentic AI Strategy. Graduated autonomy maturity model; only 11% of organisations in active agentic production; 42% still developing strategy roadmap. <a href="https://www.deloitte.com/us/en/insights/topics/technology-management/tech-trends/2026/agentic-ai-strategy.html">deloitte.com</a></li><li>McKinsey — State of AI Trust in 2026: Shifting to the Agentic Era. Governance and execution gap; approximately 30% of organisations at AI strategy maturity level three or above. <a href="https://www.mckinsey.com/capabilities/tech-and-ai/our-insights/tech-forward/state-of-ai-trust-in-2026-shifting-to-the-agentic-era">mckinsey.com</a></li></ol><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=f57980e195c0" width="1" height="1" alt=""><hr><p><a href="https://medium.com/naturally-stupid/the-intelligence-decade-what-the-6g-ai-convergence-actually-means-for-business-leaders-f57980e195c0">The Intelligence Decade: What the 6G-AI Convergence Actually Means for Business Leaders</a> was originally published in <a href="https://medium.com/naturally-stupid">Naturally Stupid</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[The Story of Your Life: A Short Story That Stays With You]]></title>
            <link>https://medium.com/@mareers/the-story-of-your-life-a-short-story-that-stays-with-you-90f4bfffc2ba?source=rss-d82d3ee616ff------2</link>
            <guid isPermaLink="false">https://medium.com/p/90f4bfffc2ba</guid>
            <category><![CDATA[book-review]]></category>
            <category><![CDATA[linguistics]]></category>
            <category><![CDATA[short-story]]></category>
            <category><![CDATA[reading]]></category>
            <category><![CDATA[science-fiction]]></category>
            <dc:creator><![CDATA[Sreeram Narasimhan]]></dc:creator>
            <pubDate>Sun, 05 Apr 2026 18:19:53 GMT</pubDate>
            <atom:updated>2026-04-05T18:19:53.682Z</atom:updated>
            <content:encoded><![CDATA[<p>In 1770, Captain Cook’s sailors pointed at the strange animals hopping around with their young in pouches and asked an aboriginal Australian what they were called. The man replied: “Kanguru.” The sailors took it as the name. It wasn’t until later that they discovered it meant “What did you say?”</p><p>Ted Chiang uses this anecdote in “The Story of Your Life” and it is the perfect summary of everything the story is exploring: the gap between two languages, two modes of perception, and what happens when beings try to communicate across that gap.</p><p>I came across this book through the <a href="https://www.goodreads.com/group/show/1187035-the-short-story-club">Goodreads Short Story Club</a>. It is a novella, not a novel, and it moves quickly. But it left a surprising amount behind.</p><p><em>(If you haven’t read it yet, the </em><a href="https://waldyrious.neocities.org/ted_chiang/story-of-your-life"><em>full text is available here</em></a><em>. Short read. Worth going in knowing nothing more.)</em></p><h3><strong>What the story is about</strong></h3><p>The narrator is Louise Banks, a linguist tasked with making contact with alien beings called heptapods who have arrived on Earth. The story shifts between two timelines: the official, present-tense work of decoding the heptapod language, and personal fragments addressed to Louise’s daughter, toggling between past and future, sometimes in the same sentence.</p><p>The story opens on a tragic note: Louise addressing her daughter about an evening they will never share together. It is a heavy way to begin. But the non-linear arc, moving between past and future, sometimes in the same sentence, makes sense of that heaviness by the end.</p><p>The non-linear structure is not just stylistic. It is the point. The more Louise learns the heptapod language, the more her own perception of time begins to shift.</p><h3><strong>A deep dive into language</strong></h3><p>This is where the story surprised me. I went in expecting a first-contact science fiction story. What I got was a fairly serious exploration of linguistic theory.</p><p>The difference between graphemes and phonemes: the written symbol versus the spoken sound. The distinction between logographic writing, where each symbol represents a whole word or concept, and glottographic writing, where symbols represent sounds. Semasiography versus orthography. The heptapods write in semagrams: dense, logographic symbols that carry meaning all at once, not sequentially.</p><p>Which leads to a structural consequence that I found genuinely interesting: to write in this system, you have to know the full statement before you put it down. You cannot begin a sentence and feel your way to the end. The meaning has to exist complete in your mind first.</p><p>I initially read this as a limitation. But from a different perspective, it is exactly consistent with how the heptapods perceive time: all at once, not in sequence. Their writing system is not a constraint. It is a reflection of how they already think.</p><h3><strong>The physics anchor: Fermat’s principle</strong></h3><p>The linguistic ideas connect to physics through Fermat’s principle of least time, which describes how light does not travel by any random path but by the path that minimises time. Chiang uses this to draw out a larger distinction: the causal formulation of physics (cause leads to effect, in order) versus the teleological formulation (everything moves toward an end-state that is already known).</p><p>Humans perceive time causally. Heptapods perceive it teleologically. The physics is not decorative. It is doing real work in the story.</p><h3><strong>A quiet parallel to Lewis Carroll</strong></h3><p>There is a thread running through the story that connects to Lewis Carroll’s “Through the Looking Glass”: the idea of looking back in order to look forward.</p><p>In Carroll’s world, the Looking Glass reverses everything. You move by going the opposite direction. Memory works the wrong way around. It is played for absurdity and wit, but the underlying intuition is not absurd: that time’s direction might be a matter of how you are standing, not how things actually are.</p><p>Ted Chiang takes the same inversion and grounds it in something more rigorous. Louise Banks, through her contact with the heptapods, develops what Chiang calls a teleological perception of time: a view that is holistic, where past and future are not two separate territories but a single landscape she can see all at once. She is not remembering the past or anticipating the future. She holds both simultaneously, the way the Looking Glass holds an image that is real and reversed at the same time.</p><p>Carroll turns time inside out for comedy. Chiang turns it inside out to ask what it would actually feel like from inside. That is the difference between the two books, and also why they sit so naturally alongside each other.</p><h3><strong>The passage I keep returning to</strong></h3><blockquote><em>When the ancestors of humans and heptapods first acquired the spark of consciousness, they both perceived the same physical world, but they parsed their perceptions differently; the worldviews that ultimately arose were the end result of that divergence. Humans had developed a sequential mode of awareness, while heptapods had developed a simultaneous mode of awareness. We experienced events in an order, and perceived their relationship as cause and effect. They experienced all events at once, and perceived a purpose underlying them all. A minimizing, maximizing purpose.</em></blockquote><p>This is the heart of the story. Everything else, the linguistics, the physics, the shifting timelines, builds toward this.</p><h3><strong>The ending</strong></h3><p>The reasons for the heptapods’ arrival and departure are never fully explained. Louise appears to have gained something from the contact: the ability to perceive time the way the heptapods do, knowing future moments the way you know memories. Whether that is a gift or a burden the story leaves open.</p><p>The last lines sit with that ambiguity:</p><blockquote><em>From the beginning I knew my destination, and I chose my route accordingly. But am I working toward an extreme of joy, or of pain? Will I achieve a minimum, or a maximum?</em></blockquote><p>A story that knows how it ends and still will not tell you.</p><p>The reasons for the heptapods’ visit, and why they left without explanation, remain open. Did they come to give Louise this ability, or was that just a side effect of contact? Did they achieve whatever they came for? Did they already know they would leave before they arrived?</p><p>I have my own read on it, but I am curious: if you have read this story, what do you think actually happened? Leave your interpretation in the comments.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*BmF4EoctY_R4p9KybjHl8A.png" /><figcaption>Image generated with Gemini</figcaption></figure><p><em>“The Story of Your Life” is part of Ted Chiang’s collection “Story of Your Life and Others.”</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=90f4bfffc2ba" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[I Let an AI Manage My Apartment Accounts. Here’s What Actually Happened.]]></title>
            <link>https://medium.com/naturally-stupid/i-let-an-ai-manage-my-apartment-accounts-heres-what-actually-happened-5f6d8507a4d9?source=rss-d82d3ee616ff------2</link>
            <guid isPermaLink="false">https://medium.com/p/5f6d8507a4d9</guid>
            <category><![CDATA[productivity]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[claude-cowork]]></category>
            <category><![CDATA[accounting]]></category>
            <category><![CDATA[personal-finance]]></category>
            <dc:creator><![CDATA[Sreeram Narasimhan]]></dc:creator>
            <pubDate>Fri, 27 Mar 2026 04:13:38 GMT</pubDate>
            <atom:updated>2026-03-28T03:33:01.753Z</atom:updated>
            <content:encoded><![CDATA[<h3>I Let Claude Cowork Manage My Apartment Accounts Excel. Here’s What Actually Happened.</h3><p>I manage the accounts for my apartment association. I did not volunteer for this. It happened the way most thankless jobs happen — by being the person who showed up, cared slightly more than average, and said “fine, I’ll do it” one too many times.</p><p>The job involves fifteen flats, monthly maintenance, electricity bills, security salaries, plumbing emergencies, generator diesel, water tankers — all of it flowing through a Google Sheet that I maintain, reconcile against our bank statements, and occasionally stare at with mild despair.</p><p>I’ve been doing this for three years. It works. But it is deeply, relentlessly tedious.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*iaSYI-C-SzdoC0br" /></figure><h3>The App I Was Looking For (That Didn’t Quite Exist)</h3><p>Before I landed here, I went looking for something purpose-built. Platforms like Apartment Adda promise to do exactly this — manage society accounts, track maintenance, send reminders, even reconcile against bank feeds. And they do. But when I sat down and mapped what I actually needed against what each plan offered and what it cost, the math didn’t add up. I have fifteen flats. The basic tier was too basic. The plan with the features I needed was priced for a complex ten times our size. I wasn’t going to spend that kind of money to automate something I was already doing, however painfully.</p><p>So I stayed with the spreadsheet. And the manual reconciliation. And the mild despair.</p><p>Then I got access to Claude Cowork — Anthropic’s desktop tool that lets you work with files on your computer using an AI — and I thought: <em>let me just try this with the accounts.</em></p><h3>The Task</h3><p>The job was straightforward on paper. I had a bank export covering December 2025 to March 2026 — about 60 transactions. I needed to match each one against the right flat, add it to the accounts spreadsheet in the correct date order, and calculate the running balances. Then there were a handful of cash expenses — EB bills, diesel for the generator, water tankers, overhead tank cleaning — that needed to be added too.</p><p>In the past, this would take me an evening. Cross-referencing bank references with payer names, splitting arrear payments across months, making sure the formulas didn’t break. Manageable, but the kind of work that makes you feel like you’re slightly wasting your life.</p><p>I opened Cowork, pointed it at my accounts folder, and described the task.</p><h3>What It Got Right</h3><p>The first thing that surprised me was how well it handled the mapping problem. Our bank statement references look like <em>NEFT/HDFC0000234/HDFC/NAME</em> — which is not exactly how anyone’s name appears in normal life. I have fifteen flats with fifteen different payment patterns, some of them irregular, some switching banks mid-year, some paying arrears in chunks.</p><p>I asked Claude to figure out which reference corresponded to which flat. It went through nearly 900 rows of transaction history, identified the patterns, flagged the ambiguities — two flats with similar names in the reference, one flat that changed payers in November — and built a reference map. Things I would have done by memory and gut feel, it did by reading the data.</p><p>Then it started adding entries. Each transaction in the right row, in date order, with the correct formula for running bank balance and cash balance. I’d ask it to place an entry, it would tell me exactly which row it went to and what the balance looked like. I’d cross-check against the bank statement. It matched.</p><p>We ran this checkpoint five times across the period. Every time, the bank balance it calculated matched what the bank said it should be. That’s when I started actually trusting it.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/900/1*nCerjpUC2NMm3KZYHUB1Wg.jpeg" /></figure><h3>What Went Wrong (And What That Looked Like)</h3><p>Here’s the part I want to be honest about, because most AI write-ups skip this.</p><p>At one point, it tried to rewrite a section of the spreadsheet to fix a date-ordering problem. The rewrite broke something in row 6 — a hardcoded opening balance that it replaced with a formula referencing an empty cell. The cascade was immediate: nearly 1,900 rows of <em>#VALUE!</em> errors, every balance in the spreadsheet destroyed.</p><p>It caught it. Explained exactly what had happened — the formula it wrote assumed row 6 followed the same pattern as every other row, but row 6 is special, it’s hardcoded, and it hadn’t accounted for that. It restored from the original file, rebuilt the entries cleanly, and we moved on.</p><p>Would I have noticed that error if I’d been doing it manually? Probably. But it would have taken me twenty minutes of confused scrolling to understand what I’d done wrong. Claude had a diagnosis in seconds.</p><p>There was also a file corruption incident — a repair prompt that Excel kept throwing when opening the file. We spent time trying to fix it by manipulating the zip structure of the xlsx file (Excel files are actually zips inside, apparently), broke it completely once, restored it, and eventually got it to a state where the repair prompt appeared but the file opened fine. I decided to live with it. Sometimes that’s the right answer.</p><h3>The Conversation I Didn’t Expect to Have</h3><p>The most interesting moment in the whole session wasn’t about data entry. It was about accounting.</p><p>My cash balance had drifted deeply negative because large expenses like EB bills and generator diesel had been recorded as Cash payments. I’d been doing it that way because I pay these personally from my own account and don’t always reimburse myself immediately. The cash account was effectively a running tab of money I was owed.</p><p>Claude spotted it, showed me the analysis — here’s when it went negative, here are the entries that caused it, here are two scenarios for how to fix it — and then asked a question I hadn’t quite articulated to myself: <em>were these actual cash payments, or personal payments you’re tracking for later reimbursement?</em></p><p>That’s an accounting distinction. Not a complicated one, but one that had practical consequences for how the Payment Status sheet worked. The sheet had formulas tracking outstanding dues by flat. If the payment mode was wrong, the formula missed the entry and the dues were understated.</p><p>We worked through it together. Some entries stayed as Cash. Some moved to Individual. We drafted a WhatsApp message to the residents’ group requesting reimbursement for the personal payments. One of my own entries from December — a bank reimbursement I’d already processed — had no flat assigned, so the formula was silently ignoring it. I fixed it manually; the outstanding dues dropped automatically.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*6TActCwdhMPhKNVsfKVuPQ.png" /></figure><p>I’ve been managing these accounts for three years. I had never quite seen that clearly how the Payment Status sheet actually worked.</p><h3>What Remains Manual — And Why I’m Fine With That</h3><p>Here’s something I haven’t seen written about enough: there are parts of this I deliberately kept manual.</p><p>I don’t have the nerve to link Claude Cowork directly with the Google Sheets where I manage the accounts. Instead, I worked through this by starting with a downloaded file in xlsx format and asking Claude to work on that. Once I was happy with all the edits, I transferred the contents back into the parent Google Sheet myself. This way, I get to see every transaction and watch how the numbers move before anything gets committed.</p><p>Perhaps I don’t need to be so cautious. But I feel better having that control with me still.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/968/1*qtLtbhhT3KXpboPeSbfCkg.png" /></figure><p>There’s another layer to this. I am an Excel person. I genuinely enjoy tinkering with formulas, watching them break, figuring out why. It’s how I’ve learned most of what I know about how this spreadsheet actually works. Handing that over entirely would mean giving up something I didn’t know I valued until I thought about giving it up.</p><p>This made me think about why a lot of engineers are also not fully automating their coding workflows even when the tools are clearly capable. Maybe people need to reach a certain level of fluency in a domain — enough to know what good looks like, enough to catch the error in row 6 — before they feel ready to fully devolve control to AI. The tool works best when you understand what it’s doing.</p><h3>Was It Faster? Did It Replace the Manual Process?</h3><p>Certainly faster, yes. A session that would have taken me an evening took a few hours of back-and-forth conversation.</p><p>Would it completely replace the manual process? Probably not entirely, and I’m not sure I want it to. But what it clearly did was take away the drudge work. The part that requires sitting still for a long time, not losing focus, cross-referencing one number against another, being precise in a way that is deeply unglamorous — that part got considerably lighter.</p><p>Reconciliation and accurate bookkeeping demand a level of concentration and personal probity that is hard to sustain when you’re also, say, the person who has to message everyone about the broken lift. AI has genuinely helped with that. I can look at a hundred unreconciled rows now without the usual low-grade dread, because I know the process I’m going to use and I know where to run my checks. That peace of mind is underrated.</p><h3>What It Feels Like to Use</h3><p>People ask about AI tools as though the question is “does it work?” But that’s not quite the right question. The right question is: <em>what kind of working relationship does it create?</em></p><p>This felt like working with a very capable, very patient colleague who has read every document in the room and remembers all of it. It doesn’t get bored. It doesn’t resent being asked to go back and check something. When it makes a mistake — and it does — it explains the mistake clearly and fixes it without defensiveness.</p><p>What it can’t do is replace judgment. Every time there was ambiguity — which flat does this payment belong to? Should this be Cash or Individual? What date should I use for a payment made “last week”? — it asked. Those were exactly the right moments to ask.</p><p>I also found myself being more precise about my own accounting logic than I usually am, because I had to explain it to something that would act on what I said. Vagueness is the enemy of good accounts, and having to articulate my reasoning out loud — or in text — forced a clarity I don’t always bring when I’m just quietly doing it myself.</p><h3>Where I’m Taking This Next</h3><p>The backlog is cleared. Now the question is whether I can turn this into a repeatable monthly rhythm rather than a periodic catch-up exercise.</p><p>The plan is simple: after the first week of each month — by which point most maintenance payments would have come in and salaries transferred — I download the bank statement, hand it to Claude Cowork along with the working Excel file, add any cash expenses for the month, validate the output, and copy the verified entries into the master Google Sheet. Rinse and repeat.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/900/1*KSgQd1vy_YMyq5YnZgX1NA.jpeg" /></figure><p>Each cycle should get faster. The flat-to-payer mapping is established. The patterns are known. The only genuine work left is the judgment calls — the unusual payments, the ambiguous references, the entries that don’t fit a template. Everything else is process.</p><h3>Could This Be More Automated? Should It?</h3><p>This is the question I keep coming back to, and I don’t have a clean answer.</p><p>There are things I could do to reduce the manual steps further. Working more directly with Google Sheets instead of a downloaded Excel copy. Triggering the bank statement download automatically. Building a more structured prompt that handles the full monthly cycle in one go. The pieces are probably all there.</p><p>But I’m genuinely uncertain about how far to take it. The manual transfer step — copying entries from Excel back into Google Sheets — is where I see every transaction one more time before it becomes permanent. That review catches things. Not always, but sometimes. Removing it means trusting the process completely, and I’m not sure I’m there yet.</p><p>The question I keep sitting with: <em>where does the human judgment actually need to live in a process like this, and what happens to the quality of the output when you remove it?</em></p><p>I’d love to hear how others have thought about this — especially if you’re managing something similar, whether it’s society accounts, small business books, or just personal finance that got complicated. What have you automated, what have you kept manual, and what broke when you got the balance wrong?</p><p><em>Sreeram Narasimhan lives in Chennai and manages apartment accounts nobody asked him to manage. He writes at </em><a href="https://medium.com/naturally-stupid"><em>Naturally Stupid</em></a><em>.</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=5f6d8507a4d9" width="1" height="1" alt=""><hr><p><a href="https://medium.com/naturally-stupid/i-let-an-ai-manage-my-apartment-accounts-heres-what-actually-happened-5f6d8507a4d9">I Let an AI Manage My Apartment Accounts. Here’s What Actually Happened.</a> was originally published in <a href="https://medium.com/naturally-stupid">Naturally Stupid</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[I Would Prefer Not To]]></title>
            <link>https://medium.com/@mareers/i-would-prefer-not-to-1de10eea3067?source=rss-d82d3ee616ff------2</link>
            <guid isPermaLink="false">https://medium.com/p/1de10eea3067</guid>
            <category><![CDATA[future-of-work]]></category>
            <category><![CDATA[history]]></category>
            <category><![CDATA[writing]]></category>
            <category><![CDATA[technology]]></category>
            <category><![CDATA[literature]]></category>
            <dc:creator><![CDATA[Sreeram Narasimhan]]></dc:creator>
            <pubDate>Wed, 18 Mar 2026 16:59:57 GMT</pubDate>
            <atom:updated>2026-03-18T17:57:23.397Z</atom:updated>
            <content:encoded><![CDATA[<p>Ask a man why he has stopped working, and he will tell you. Ask him again, and he will elaborate. Ask Bartleby — the quiet copyist at the centre of <a href="https://www.gutenberg.org/ebooks/11231">Herman Melville’s 1853 story</a> — and he will look at you, and say:</p><p><em>I would prefer not to.</em></p><p>This is not defiance. There is no anger in it, no ideology, no grievance filed. Bartleby was once the very diligent Scrivener in a Wall Street law office, copying documents by hand day after day, with the focus that impressed even his employer — especially considering this particular employer employed people who came with their own quirks that made them quite unproductive at different times of the day. And then, gradually, he simply stopped. The extra tasks first. Then the copying. Then eating. The lawyer who employs him tries reason, tries kindness, tries alternatives. Bartleby receives all of it the same way. Quietly. Without malice.</p><p>Ask him what is wrong, and he prefers not to say. He was not the last one to feel that way.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*ADTB0ZGd00qi2l5-T0m8Tg.png" /><figcaption>I would prefer not to</figcaption></figure><p><strong>1850s. The Scrivener.</strong></p><p>He spent his days bent over parchment, copying words that were not his, in an office that faced a wall. His hand knew the work before his mind did.</p><p><em>I would prefer not to.</em></p><p><strong>1860s. The Telegraph Operator.</strong></p><p>His fingers tapped out the pulse of a nation — wars, deaths, markets, marriages — in a language of dots and dashes only he could read at speed. The messages passed through him like electricity, and left just as quickly.</p><p><em>I would prefer not to.</em></p><p><strong>1900s. The Horse Carriage Driver.</strong></p><p>He knew these streets longer than the men who owned them — every cobblestone, every corner, every passenger. Then the motorcars came, and they asked him to be grateful for the transition.</p><p><em>I would prefer not to.</em></p><p><strong>1910s. The Typesetter.</strong></p><p>He arranged individual letters by hand, one by one, so that the world could read its morning news. It took years to master. It took one machine to make it history.</p><p><em>I would prefer not to.</em></p><p><strong>1920s. The Lift Operator.</strong></p><p>He stood for eight hours in a box the size of a wardrobe, the first face people saw and the last when they left. Nobody thought about what it cost to smile that many times.</p><p><em>I would prefer not to.</em></p><p><strong>1930s. The Stenographer.</strong></p><p>She captured every word spoken in boardrooms and courtrooms at impossible speed, invisible in the room, indispensable to the record. The men whose words she preserved rarely learned her name.</p><p><em>I would prefer not to.</em></p><p><strong>1940s. The Telephone Operator.</strong></p><p>She was the calm voice that connected a world coming apart at the seams, routing calls through chaos, keeping the lines of human contact open. When the war ended, they automated the switchboard and thanked her for her service.</p><p><em>I would prefer not to.</em></p><p><strong>1950s. The Double-Entry Bookkeeper.</strong></p><p>She could find an error in ten thousand entries without breaking a sweat: decades of precision that bordered on art. They brought in a machine that did it faster, then asked her to teach it her instincts.</p><p><em>I would prefer not to.</em></p><p><strong>1960s. The Bank Teller.</strong></p><p>He knew every customer by name, every account by instinct, every transaction balanced to the penny by hand. Then came the ATM, and then the app, and then the question of what exactly he was still there for.</p><p><em>I would prefer not to.</em></p><p><strong>1970s. The Assembly Line Worker.</strong></p><p>The same motion, ten thousand times a day — his hands carried the history of everything that rolled, flew, or ran in this country. The line sped up. Management called it efficiency.</p><p><em>I would prefer not to.</em></p><p><strong>1980s. The Hotel Doorman.</strong></p><p>He remembered names, preferences, the small dignities that made a guest feel seen — thirty years at the same post through every renovation and rebrand. They installed an automated door and told him his new role was “guest experience facilitator.”</p><p><em>I would prefer not to.</em></p><p><strong>1990s. The Data Entry Clerk.</strong></p><p>She typed the entire paper world into the digital one, patient, precise, the invisible bridge between what was and what was coming. Once the bridge was built, nobody needed the builder anymore.</p><p><em>I would prefer not to.</em></p><p><strong>2010s. The Content Moderator.</strong></p><p>She reviewed what human beings do to each other when they think no one is watching — violence, cruelty, things without names yet — so your feed stayed clean. She was told to take breaks. She was not told how to put herself back together after.</p><p><em>I would prefer not to.</em></p><p><strong>2020s. The Software Engineer.</strong></p><p>He was told he was building the future, and for a while the problems were interesting, the craft was his, the work carried his fingerprints. Then the tools arrived that could code faster, document cleaner, and debug more patiently — and they asked him to justify why they still needed him at all.</p><p><em>I would prefer not to. But I am not particular.</em></p><p>Ah Bartleby! Ah humanity!</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=1de10eea3067" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Exam prep in 2026: Me, my kid, Claude, and a school textbook.]]></title>
            <link>https://medium.com/naturally-stupid/exam-prep-in-2026-me-my-kid-claude-and-a-school-textbook-f84ff8065654?source=rss-d82d3ee616ff------2</link>
            <guid isPermaLink="false">https://medium.com/p/f84ff8065654</guid>
            <category><![CDATA[parenting]]></category>
            <category><![CDATA[productivity]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[education]]></category>
            <category><![CDATA[claude]]></category>
            <dc:creator><![CDATA[Sreeram Narasimhan]]></dc:creator>
            <pubDate>Sun, 15 Mar 2026 15:31:40 GMT</pubDate>
            <atom:updated>2026-03-15T15:31:40.065Z</atom:updated>
            <content:encoded><![CDATA[<p>Exam season arrived, and with it the usual question: how do you turn three hundred pages of a school textbook into something a kid can actually revise from?</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*9gcH9MLLe4WwUGMr34Vjlg.png" /></figure><p>My initial idea was to use Claude for Chrome to take screenshots from the eBook and convert them into PDF and then feed it into NotebookLM to generate answers and key learning materials.</p><p>While Claude Pro was able to do the task of taking screenshots and converting them into PDFs, it was becoming too cumbersome. Also, the output was not of great quality, and then we have the problem of NotebookLM or any other tool needing to read a poor quality screenshot PDF and extract the text again and then go through further process steps.</p><p>Then I thought, why not make this simpler? Why should I use two LLMs? I asked Claude if it would be able to read the contents of a chapter and then help me answer the end-of-chapter questions and create some study material for exam revision. And voila! This is what I did with Claude. Claude gave me a plan with the following steps and then it was quite cool in terms of creating an output in a format that I wanted.</p><h3>Claude’s Plan</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*mLEC7-1EKMc3gVUUKy1Uew.png" /><figcaption>Claude reading through the textbook chapter</figcaption></figure><p><strong>Approach:</strong><br>1. Navigate to Chapter by clicking forward from the current page<br>2. Read all pages of Chapter by zooming into each page for clear text<br>3. Identify and read all end-of-chapter questions<br>4. Answer all questions based solely on Chapter content<br>5. Format everything as exam study material matching the style of previous chapters<br>6. Export the study material as a downloadable RTF/Word document</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*E8sb3zt6cM_2VLp36gmEEw.png" /><figcaption>Claude’s formatted study material output</figcaption></figure><p>So what are the downsides? Every cool idea will definitely have some downsides to it. In this case I can see a couple:</p><ol><li>The process is quite slow. It’s not as fast as quickly manually taking screenshots and moving on. But how does it matter if I can initiate a process on the browser using Claude and then let it slowly trudge through this and get me a clean output? It still works better than manually doing this faster.</li><li>I’m unable to do this repeatedly and consistently. The same prompt from one time to the next gives different results. This creates a serious issue of consistency in the output, something very important when working on such content.</li><li>This process also seems to make me hit the limits very quickly. Perhaps I should try with Haiku but I am not sure on the accuracy of the output so I don’t want to take a chance at the moment. Perhaps next time around I’ll try Haiku and see if I can do this without hitting the limit.</li><li>There is a good chance of hallucination. So always need to follow the mantra of “Trust but verify.”</li></ol><figure><img alt="" src="https://cdn-images-1.medium.com/max/454/1*VpitVbbJtJTY7mwPjhOFyw.png" /><figcaption>Claude hitting limits — painful wait to reset</figcaption></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/437/1*1hZ4zyceW0y4I2hPxkgS4Q.png" /><figcaption>Hallucination or inconsistency in the output — always “trust, but verify”</figcaption></figure><p>For parents wanting to experiment with this: always remember that adage, “trust but verify.” It’s important to try different options and see what suits your requirements best. There is no one size fits all solution here. What worked for one chapter might not work the next time, and NotebookLM might handle some of these issues better. More on that in a subsequent post.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=f84ff8065654" width="1" height="1" alt=""><hr><p><a href="https://medium.com/naturally-stupid/exam-prep-in-2026-me-my-kid-claude-and-a-school-textbook-f84ff8065654">Exam prep in 2026: Me, my kid, Claude, and a school textbook.</a> was originally published in <a href="https://medium.com/naturally-stupid">Naturally Stupid</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[75 Years is a Horizon. 4,000 Weeks is a Countdown.]]></title>
            <link>https://medium.com/@mareers/75-years-is-a-horizon-4-000-weeks-is-a-countdown-7167eb3ae0d5?source=rss-d82d3ee616ff------2</link>
            <guid isPermaLink="false">https://medium.com/p/7167eb3ae0d5</guid>
            <category><![CDATA[personal-development]]></category>
            <category><![CDATA[life-lessons]]></category>
            <category><![CDATA[self-improvement]]></category>
            <category><![CDATA[philosophy]]></category>
            <category><![CDATA[mathematics]]></category>
            <dc:creator><![CDATA[Sreeram Narasimhan]]></dc:creator>
            <pubDate>Fri, 13 Mar 2026 16:38:32 GMT</pubDate>
            <atom:updated>2026-03-13T16:38:32.987Z</atom:updated>
            <content:encoded><![CDATA[<p>I’m always fascinated by numbers. I used to think about numbers like 36 as quite unique — simply because it’s 3² × 2², the product of two numbers each raised to a power. So it’s no surprise that when I recently turned 42, my first thought was: “What’s so unique about this number?”</p><p>Well, 41 was unique because it was prime. And 40 was obvious — a multiple of 10, the threshold into a new decade. But 42? It’s 6 × 7. Two successive numbers multiplied together. The last time I celebrated an age like that was when 12 years ago — that’s 5 × 6. And the next time? That’ll be 14 years later, when it’s 7 × 8. There’s something quietly wonderful about the way these numbers play out, each one asking to be understood on its own terms.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*3nT8dh1KsvN59qOAUml_tw.png" /></figure><h3>So What Kind of Age Is Yours?</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*hea6LH0PdXY3_rhF4A66VQ.png" /><figcaption><em>Find your age. What colour are you?</em></figcaption></figure><p>Since we’re here, let’s go all the way. Every age from 1 to 90 belongs to at least one interesting mathematical family. Here’s a completely unnecessary, and yet oddly satisfying, taxonomy.</p><p><strong>The Prime of Your Life:</strong>These are the years that stand completely alone. Indivisible. If you’re 2, 3, 5, 7, 11, 13, 17, 19, 23, 29, 31, 37, 41, 43, 47, 53, 59, 61, 67, 71, 73, 79, 83, or 89 — congratulations. You are, quite literally, in your prime.</p><p><strong>The Staircase Years: </strong>Ages that are the product of two consecutive numbers (n × (n+1)). Each one a natural step up: 2, 6, 12, 20, 30, 42, 56, 72, 90. I may have mentioned that 42 belongs here. Just once or twice.</p><p><strong>The Square Years: </strong>When everything lines up perfectly. 1, 4, 9, 16, 25, 36, 49, 64, 81. The years that, in a purely mathematical sense, have it all figured out.</p><p><strong>The Fibonacci Years: </strong>The ages where nature itself seems to be paying attention. Each number in this sequence is the sum of the two before it: 1, 2, 3, 5, 8, 13, 21, 34, 55, 89. If you’re at one of these ages, you are part of a pattern found in sunflowers, spiral galaxies, and nautilus shells. No pressure.</p><p><strong>The Round Numbers: </strong>10, 20, 30, 40, 50, 60, 70, 80, 90. The ones that get the big parties, the milestone dinners, and the unsolicited life advice. Whether they deserve the fuss is debatable. They get it anyway.</p><p>And then there are the overachievers — ages that fall into more than one category. Age 89 is both prime <em>and</em> Fibonacci. Age 64 is a perfect square, a power of 2, <em>and</em> a perfect cube. Some ages simply refuse to be ordinary.</p><p>So why talk about age as a number at all?</p><p>Partly because I’ve always been this way — questions about existence, about what comes before and after our time here, have been with me since I was nine or ten. This isn’t a mid-life reckoning. It’s just a recurring conversation I have with myself, and 42 gave me a fresh reason to have it again.</p><p>But more immediately, it was one single idea that stopped me.</p><p>An average human has around <strong>4,000 weeks</strong> to live. It’s the premise behind Oliver Burkeman’s book <a href="https://www.oliverburkeman.com/fourthousandweeks"><em>Four Thousand Weeks: Time Management for Mortals</em></a> — and when I came across it, something shifted.</p><p>75 years feels long. Manageable. Even generous, in the way big numbers can feel. But 4,000 weeks — the same span, just measured differently — felt suddenly tight. The unit of measure does something to your perception. 75 years is a horizon. 4,000 weeks is a countdown.</p><p>Burkeman’s argument isn’t really about getting more done. It’s about deciding what to leave undone. That idea sat with me differently once I placed it next to what Rajinikanth says about the fifth chapter of eight.</p><p>One frame I’ve always loved comes from the film <em>Basha</em> — the song <a href="https://www.youtube.com/watch?v=MSSUNGc-KKw"><em>Ra ra Ramaya</em></a>, in which Superstar Rajinikanth divides a life into chapters of eight years each. By that measure, at 42, I’m in the fifth chapter. And what does he say about the fifth chapter? <em>“Aindam ettil serkadhudhu selvammum alla”</em> — roughly: <em>in the fifth eight, don’t chase wealth either.</em> It’s an inward turn. A call to look within rather than accumulate. Selfish, perhaps — but not wrongly so.</p><iframe src="https://cdn.embedly.com/widgets/media.html?src=https%3A%2F%2Fwww.youtube.com%2Fembed%2FMSSUNGc-KKw%3Ffeature%3Doembed&amp;display_name=YouTube&amp;url=https%3A%2F%2Fwww.youtube.com%2Fwatch%3Fv%3DMSSUNGc-KKw&amp;image=https%3A%2F%2Fi.ytimg.com%2Fvi%2FMSSUNGc-KKw%2Fhqdefault.jpg&amp;type=text%2Fhtml&amp;schema=youtube" width="854" height="480" frameborder="0" scrolling="no"><a href="https://medium.com/media/d518c41f02688fab78a43763f85d8643/href">https://medium.com/media/d518c41f02688fab78a43763f85d8643/href</a></iframe><p>Then there’s the other frame, which goes deeper. The song <a href="https://www.youtube.com/watch?v=Kh_BX1poApo"><em>Saga Varam</em></a> from <em>Uttama Villain</em> holds a line that always hits me: <em>“Vidhaithidummei pol oru uyirai uyirtu vilangum”</em> — <em>like planting a seed, raise a life that shines.</em> It’s about legacy. About what we leave behind when our number finally runs out. And the answer the song offers is quietly profound: perhaps the only thing worth leaving behind to speak for us is our children.</p><iframe src="https://cdn.embedly.com/widgets/media.html?src=https%3A%2F%2Fwww.youtube.com%2Fembed%2FKh_BX1poApo%3Ffeature%3Doembed&amp;display_name=YouTube&amp;url=https%3A%2F%2Fwww.youtube.com%2Fwatch%3Fv%3DKh_BX1poApo&amp;image=https%3A%2F%2Fi.ytimg.com%2Fvi%2FKh_BX1poApo%2Fhqdefault.jpg&amp;type=text%2Fhtml&amp;schema=youtube" width="854" height="480" frameborder="0" scrolling="no"><a href="https://medium.com/media/4b91e923324c2c53ca2acdc292129f46/href">https://medium.com/media/4b91e923324c2c53ca2acdc292129f46/href</a></iframe><p>Two songs, two very different answers to the same question.</p><p>I don’t think there’s a single right way to read your age. But there is one quote — often attributed to Gandhi — that I keep returning to:</p><blockquote>“Live as if you were to die tomorrow. Learn as if you were to live forever.”</blockquote><p>That tension, between urgency and patience, between finite weeks and infinite curiosity, feels like the most honest answer I’ve found.</p><p>So here’s what I’ll leave you with: go on, do the maths. Take your age, multiply it by 52, and subtract from 4,000. How many weeks do you have left? Sit with that number for a moment. Not with dread — but with intention.</p><p>What are you going to do with yours?</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=7167eb3ae0d5" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Twenty Logos, Twenty Google Searches. I Finally Stopped.]]></title>
            <link>https://medium.com/naturally-stupid/twenty-logos-twenty-google-searches-i-finally-stopped-8532fc3ca607?source=rss-d82d3ee616ff------2</link>
            <guid isPermaLink="false">https://medium.com/p/8532fc3ca607</guid>
            <category><![CDATA[workflow]]></category>
            <category><![CDATA[design]]></category>
            <category><![CDATA[productivity]]></category>
            <category><![CDATA[ai-tools]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <dc:creator><![CDATA[Sreeram Narasimhan]]></dc:creator>
            <pubDate>Thu, 12 Mar 2026 07:41:54 GMT</pubDate>
            <atom:updated>2026-03-12T07:41:54.407Z</atom:updated>
            <content:encoded><![CDATA[<p>Every time I needed a company logo for a presentation, I’d do the same thing. Open Google, type “[Company name] logo PNG”, right-click, save image, repeat. Twenty logos meant twenty trips to the search bar. It worked. It was also entirely unnecessary.</p><p>I was building a slide to showcase the technical ecosystem our platform integrates with — a visual map of all the tools, services, and platforms in the mix. That’s a lot of logos. I was already halfway through my usual routine when I realised there had to be a better way.</p><p>LogoKit has a massive library of company logos available as either SVG or PNG files. Clean versions, ready to use. No hunting through image search results trying to figure out which one isn’t a blurry JPEG from 2009.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*mBmgPbzLoKbM7XusxjJ_uQ.png" /><figcaption><em>LogoKit’s homepage — a clean grid of brand logos, ready to download as SVG or PNG.</em></figcaption></figure><p>What made it even better was pairing it with Comet Browser. I navigated to LogoKit.com, told the Comet assistant which logos I needed, and it searched and downloaded the whole set. The process that used to eat up 20–30 minutes of focused clicking was done before I’d finished my coffee.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*VL6BjI4Y_6CClc9nxIOXLw.png" /><figcaption><em>The full workflow: what logo hunting looked like before, and what it looks like now with Comet + LogoKit</em></figcaption></figure><p>The free tier is also genuinely generous — no login required, and up to 300–500 logos per day. I can’t imagine hitting that limit, but it’s good to know it’s there.</p><blockquote><em>The stupid part wasn’t that I didn’t know about LogoKit, or using Comet browser to automate this task. The stupid part is how long I didn’t bother to piece one and one together.</em></blockquote><p><strong>Bottom line:</strong> If you’re regularly pulling logos for decks, proposals, or ecosystem maps, LogoKit + Comet is the combination you didn’t know you needed.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=8532fc3ca607" width="1" height="1" alt=""><hr><p><a href="https://medium.com/naturally-stupid/twenty-logos-twenty-google-searches-i-finally-stopped-8532fc3ca607">Twenty Logos, Twenty Google Searches. I Finally Stopped.</a> was originally published in <a href="https://medium.com/naturally-stupid">Naturally Stupid</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[My Son Asked Me Why We Celebrate Women’s Day. I Had No Answer.]]></title>
            <link>https://medium.com/@mareers/my-son-asked-me-why-we-celebrate-womens-day-i-had-no-answer-4afbc6dd3368?source=rss-d82d3ee616ff------2</link>
            <guid isPermaLink="false">https://medium.com/p/4afbc6dd3368</guid>
            <category><![CDATA[international-womens-day]]></category>
            <category><![CDATA[women-empowerment]]></category>
            <category><![CDATA[family]]></category>
            <category><![CDATA[parenting]]></category>
            <category><![CDATA[india]]></category>
            <dc:creator><![CDATA[Sreeram Narasimhan]]></dc:creator>
            <pubDate>Sun, 08 Mar 2026 11:45:38 GMT</pubDate>
            <atom:updated>2026-03-08T13:31:16.311Z</atom:updated>
            <content:encoded><![CDATA[<p>As I was nonchalantly wishing my mom and wife Happy Women’s Day, my son gave me a moment to pause and think.</p><p>“Is today Indira Gandhi’s birthday?”</p><p>I was caught a little dumbfounded, and so were my wife and mom. I replied, “No, and why are you asking?”</p><p>He nonchalantly responded, “If we are celebrating Children’s Day on Jawaharlal Nehru’s birthday, then perhaps we are celebrating Women’s Day on his daughter’s birthday.”</p><p>A very logical question. But my immediate thought was: why couldn’t it have been a celebration of Nehru’s wife’s birthday, or his mother’s birthday?</p><p>None of us had the right answer on why today is celebrated as Women’s Day. But this triggered my curiosity. I looked it up and shared what I found with my son and my family.</p><p>As we celebrate this day, let’s take time to understand why it is celebrated. I am not going to share this here. That would make things too easy. I will let my readers figure out the reason for celebrating Women’s Day today.</p><p>And if you do find the reason, let’s double down and celebrate the day appreciating its meaning and purpose.</p><p>Happy Women’s Day!</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*MBpQRCoLUmOUprAM.jpeg" /><figcaption>Photo by George Dolgikh from Pexels: <a href="https://www.pexels.com/photo/happy-women-s-day-box-2072166/">https://www.pexels.com/photo/happy-women-s-day-box-2072166/</a></figcaption></figure><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=4afbc6dd3368" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[I Let Claude Cowork Organise My Obsidian Vault. No Coding Involved.]]></title>
            <link>https://medium.com/naturally-stupid/i-let-claude-cowork-organise-my-obsidian-vault-no-coding-involved-627842324957?source=rss-d82d3ee616ff------2</link>
            <guid isPermaLink="false">https://medium.com/p/627842324957</guid>
            <category><![CDATA[claude-cowork]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[notetaking]]></category>
            <category><![CDATA[productivity]]></category>
            <category><![CDATA[obsidian]]></category>
            <dc:creator><![CDATA[Sreeram Narasimhan]]></dc:creator>
            <pubDate>Fri, 06 Mar 2026 12:58:53 GMT</pubDate>
            <atom:updated>2026-03-06T12:58:53.728Z</atom:updated>
            <content:encoded><![CDATA[<p>I love Obsidian. I have a premium subscription simply because it is amazing and great at creating and managing notes. The idea of having the ability to put down my random thoughts, views on books, articles, and so on was quite alluring. More of a Second Brain.</p><p>However, this comes with a lot of housekeeping. The need to create and manage structures. And consistently following them. This is very difficult to do in a hectic life journey that I ended up not using the app consistently at all.</p><p>Which is a cause of friction since here I am paying for something I love but can’t make proper use of.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*uGw2d1seGzBjuqrJif5S_A.png" /></figure><p>The problem is not Obsidian. The problem is how we think about capturing ideas.</p><p>I generally think that I can remember an idea in my mind and use it whenever I need. But that never happens. Our brains are not meant to store a lot of ideas. They are meant to think, decide, reminisce, and so on. Asking your brain to also be an archive is setting it up to fail.</p><p>So the ideas would come, and I’d tell myself I’d note it down later. And then later would become never.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*jVf3Hd3Fbt2r22EPyTLe2A.png" /><figcaption><em>My Obsidian sidebar before the cleanup. Files scattered across root, vague folder names, no frontmatter anywhere.</em></figcaption></figure><p>I like the folder structure in Obsidian. It gives a very nice way of arranging things, almost like having shelves in a cupboard to arrange books or clothes. I also understand the power of YAML, the way it helps in cataloguing content without necessarily relying only on folder structure. How this helps in giving a good summary of everything.</p><p>But doing all of these on a daily basis, when I want to spend like 5 or 10 minutes to put down a sudden thought or an idea that comes to my mind, is very difficult. The overhead kills the impulse. By the time I’ve thought about which folder this note belongs in, what tags to apply, what fields to fill in the frontmatter, the thought has already lost its energy.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*iB4P8n6B8e8yvm4Kg-Vu_A.png" /><figcaption>Five steps between having an idea and actually capturing it. With Cowork, it’s three.</figcaption></figure><p>This is what changed with <strong>Claude Cowork</strong>.</p><p>What surprised me was not Claude’s capability. I’d already been using Claude Code for a week for my non-coding explorations and had a sense of what AI assistants can do with files. What surprised me was the ease. I could focus entirely on getting my thoughts down, and the housekeeping happened around me rather than on top of me.</p><p>What I actually saw was a file move to the right folder on its own. Then YAML fields appearing in notes I had not opened in months, without me touching them. Files I had forgotten about, suddenly organised and tagged. That was the moment.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*Fx1qItEzeAQ8FAhfLARk3A.gif" /><figcaption>Claude Cowork reorganising the vault in real time. No scripts, no coding.</figcaption></figure><p>The worry about where this note will go, what I should tag it, how it connects to other things, all of that lifted. Which meant I actually wrote.</p><p>But the part I’m most optimistic about isn’t Cowork on its own. It’s the combination.</p><p>I’m also using WisprFlow to speak out my ideas rather than type them down, which saves a lot of time and effort. More importantly, WisprFlow and Obsidian are both available on Android. Which means I can simply speak out my ideas on my mobile wherever I am, into a random folder in Obsidian, and then come back to my laptop to use Cowork to put things in the right place and move forward.</p><p>Capture anywhere. Organise later. That’s the rhythm that was always missing.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*9GmmG62hMnG9rbsF8mB_0Q.png" /><figcaption><em>The full loop: speak into the phone, dump it in Obsidian, let Cowork sort the rest.</em></figcaption></figure><p>Is this setup going to last? I’m more optimistic than I’ve been about any previous system. The reason is that the effort of capture has come down to almost nothing. Speaking a thought into a phone while walking to a meeting, or just before sleeping, is a very different act from opening a laptop, creating a file, filling in a template, and tagging it correctly.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*Tw187wTpTg1Xj_3ggVVVDw.png" /><figcaption><em>The same vault, after. Clean folders, proper YAML, every file where it belongs.</em></figcaption></figure><p>The shelf doesn’t need me to arrange it anymore. I just need to put things on it.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=627842324957" width="1" height="1" alt=""><hr><p><a href="https://medium.com/naturally-stupid/i-let-claude-cowork-organise-my-obsidian-vault-no-coding-involved-627842324957">I Let Claude Cowork Organise My Obsidian Vault. No Coding Involved.</a> was originally published in <a href="https://medium.com/naturally-stupid">Naturally Stupid</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
    </channel>
</rss>