<?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 Arjita Alka Sethi on Medium]]></title>
        <description><![CDATA[Stories by Arjita Alka Sethi on Medium]]></description>
        <link>https://medium.com/@arjita_sethi?source=rss-f1bc6e9c3b01------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/0*yrZUTmXNpVgvk88A.jpg</url>
            <title>Stories by Arjita Alka Sethi on Medium</title>
            <link>https://medium.com/@arjita_sethi?source=rss-f1bc6e9c3b01------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Sat, 21 Nov 2026 13:28:52 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@arjita_sethi/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[What a room of non-technical people built with AI in one session]]></title>
            <link>https://medium.com/build-with-ai-club/what-a-room-of-non-technical-people-built-with-ai-in-one-session-009a00943e7e?source=rss-f1bc6e9c3b01------2</link>
            <guid isPermaLink="false">https://medium.com/p/009a00943e7e</guid>
            <category><![CDATA[entrepreneurship]]></category>
            <category><![CDATA[productivity]]></category>
            <category><![CDATA[small-business]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[claude]]></category>
            <dc:creator><![CDATA[Arjita Alka Sethi]]></dc:creator>
            <pubDate>Thu, 19 Nov 2026 17:31:02 GMT</pubDate>
            <atom:updated>2026-11-19T17:31:02.075Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*TbQDf7CEiWUuQHqYSriDtQ.jpeg" /></figure><p>Nobody in the room could code. There were founders, business owners, managers and moms. By the end of one session, every one of them had an AI chief of staff.</p><p>I want to describe how that happened, because it is less mysterious than it sounds.</p><p>An AI chief of staff is one project, in my case in Claude, that knows three things about you: how you work, who is who in your world, and what good looks like. That is all. It is not an app and it is not automation. It is a place where your context lives, so you stop explaining yourself from zero every time you open a chat.</p><p>We did not start by writing instructions. Most people freeze when you ask them to write a prompt, because they think there is a correct way to do it. So I turned it around and had the AI interview them. Twelve questions about their work, their week and their people. They answered the way they would answer a colleague over coffee, and the answers became the foundation.</p><p>Then came the part people skip, which is the never list. Everyone wrote down what their chief of staff must never do. Never send anything without showing me. Never guess a number. Never use a word I would not say. That short list did more for the quality of the output than any clever prompt.</p><p>Then they tried it on something real, and I asked my usual question: what did you explain twice last week? That was the first task each person handed over.</p><p>The first drafts were not good, and I told them before we started that they would not be. Early on you get around 40 percent of what you want. I compare it to a new barista. On day one they do not know your order. By week three they have it ready when you walk in, because you kept coming back and kept correcting them. People who quit at 40 percent decide AI does not work. People who stay get something that sounds like them.</p><p>What surprised the room was not the technology. It was that the work was describing themselves. People who had told me they were “not technical” turned out to be very good at this, because they know their business better than anyone.</p><p>I end the session with one rule. If this is not saving you time, there is no point. Go home, use it for a week, and count the hours.</p><p>That is what one session looks like. Nobody learned to code. Everybody left with one thing built.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=009a00943e7e" width="1" height="1" alt=""><hr><p><a href="https://medium.com/build-with-ai-club/what-a-room-of-non-technical-people-built-with-ai-in-one-session-009a00943e7e">What a room of non-technical people built with AI in one session</a> was originally published in <a href="https://medium.com/build-with-ai-club">Build With AI Club</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[You finished the AI course. Now build one thing.]]></title>
            <link>https://medium.com/build-with-ai-club/you-finished-the-ai-course-now-build-one-thing-09e6870f5406?source=rss-f1bc6e9c3b01------2</link>
            <guid isPermaLink="false">https://medium.com/p/09e6870f5406</guid>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[small-business]]></category>
            <category><![CDATA[productivity]]></category>
            <category><![CDATA[entrepreneurship]]></category>
            <dc:creator><![CDATA[Arjita Alka Sethi]]></dc:creator>
            <pubDate>Tue, 17 Nov 2026 19:46:53 GMT</pubDate>
            <atom:updated>2026-11-17T19:46:53.286Z</atom:updated>
            <content:encoded><![CDATA[<p>You did the course. You watched the modules, you saved the prompt pack, and you might even have the certificate. And on Monday morning you opened ChatGPT, typed something, got an answer that sounded like everybody else’s, and went back to doing the work by hand.</p><p>I teach AI to people who do not code, and I have taught entrepreneurship at San Francisco State for nine years. I want to tell you what I tell my students, which is that nothing is wrong with you. The course taught you what AI can do. Nobody asked what you need it to do.</p><p>You are not alone in this. Goldman Sachs found that 76 percent of small business owners use AI and only 14 percent have it in their core operations. Most people have downloaded the app and never made it part of the work.</p><p>So here is the assignment I give instead of another course. Build one thing.</p><p>Not a system and not a strategy. One thing that does one job for you every week.</p><p>Choose it by asking a single question: what did I explain twice last week? Maybe it was how you like a client email written. Maybe it was how to prepare the weekly numbers, or how to answer the same customer question for the fortieth time. If you explained it twice, you already know what good looks like, and that is exactly what AI needs from you.</p><p>Then treat it like a new hire. Give it one role. Tell it how you work, who the people are and what good looks like. Tell it what it must never do. Expect the first drafts to be around 40 percent of what you want, the same way a new barista does not get your order right on the first day. You correct it and you keep going, and it improves because you stayed with it.</p><p>And measure only one thing, which is whether it gave you time back. If this is not saving you time, there is no point.</p><p>In my Ops cohort, the first session ends with every person in the room holding an AI chief of staff they built themselves. Nobody wrote code. They answered questions about their own business, and that was enough to start.</p><p>That is the gap between knowing AI and using it. You do not close it by learning more. You close it by building one small, boring, useful thing and living with it for a few weeks.</p><p>What is the one thing you would build first?</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=09e6870f5406" width="1" height="1" alt=""><hr><p><a href="https://medium.com/build-with-ai-club/you-finished-the-ai-course-now-build-one-thing-09e6870f5406">You finished the AI course. Now build one thing.</a> was originally published in <a href="https://medium.com/build-with-ai-club">Build With AI Club</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Hand off the boring task you understand best]]></title>
            <link>https://medium.com/build-with-ai-club/hand-off-the-boring-task-you-understand-best-59ef813b3d50?source=rss-f1bc6e9c3b01------2</link>
            <guid isPermaLink="false">https://medium.com/p/59ef813b3d50</guid>
            <category><![CDATA[automation]]></category>
            <category><![CDATA[small-business]]></category>
            <category><![CDATA[entrepreneurship]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[productivity]]></category>
            <dc:creator><![CDATA[Arjita Alka Sethi]]></dc:creator>
            <pubDate>Fri, 13 Nov 2026 19:19:22 GMT</pubDate>
            <atom:updated>2026-11-13T19:19:22.776Z</atom:updated>
            <content:encoded><![CDATA[<p>Last week in the AI for Ops cohort, I put two lists side by side on the screen and asked the room which one they would rather walk out with.</p><p>The list on the left was titled “Automate everything.” It had 38 tasks on it, with an AI sales agent at the top, and by Friday nothing on it had been handed off. The list on the right was a single card for weekly client updates, the same format every Friday, written by hand every time. Both lists came from a fictional sample business I use for teaching, an interior design studio, so the numbers are illustrative, and the point lands anyway.</p><p>The left list felt ambitious. The right list actually got done, and it gave back more hours than everything on the left.</p><p>I want to make the case for the card on the right, because most of the advice about AI at work points people toward the list on the left.</p><h3>The exciting job is where people stall</h3><p>Your feed is probably full of AI sales agents, AI receptionists and businesses that claim to run themselves. Those jobs are exciting, and they are also big jobs with fuzzy edges. They need lots of judgment, and they are expensive when AI gets them wrong, because a wrong answer reaches a customer before you see it.</p><p>When people start there, they usually stall. They spend a weekend wiring something up, it does something strange on Monday, and they decide AI isn’t ready for their business. The AI usually wasn’t the problem, because the job they picked was the wrong first job.</p><h3>The job you hate most is a trap too</h3><p>The next piece of advice people hear is to start with the task they hate most. I understand the appeal, and annoyance is a great signal, because a task that annoys you is usually one you repeat. I use it as a tiebreaker in class.</p><p>The trouble is that the task we hate most is often a task we don’t fully understand. We hate the bookkeeping cleanup because the rules confuse us. We hate the proposal because every client wants something different. Hand that job to AI first and you can’t tell whether the result is right, so you end up redoing it, and you trust AI less than before you started.</p><h3>Start with what you can check in two minutes</h3><p>The first task to hand off is the boring one you understand best. You have done it so many times that you can glance at the result and know in two minutes whether it is right.</p><p>That is the real reason the weekly client update made such a good first card. The person who writes it knows exactly what a good update looks like, so checking a draft takes a minute or two per client. When AI gets something wrong, she sees it immediately. When it gets something right, she trusts it a little more. Over a few weeks, that trust is what lets her hand off the next job.</p><p>In the Job Card I teach, this is a line of its own, called the check moment. Every card says when you read the work before anything leaves. If you can’t describe your check moment, you aren’t ready to hand off the job yet.</p><h3>How to find the right first task</h3><p>The method is simple, and it takes about an hour.</p><p>Write down every task from last week with the minutes each one took. Then sort them into three piles. Repeat tasks have the same trigger, the same steps and the same result every time. Judgment tasks need you to weigh things, like pricing a project or choosing a contractor. Relationship tasks are the ones where the person wants you, like a hard conversation with a client.</p><p>Only the repeat pile goes to AI first. The other two piles aren’t off limits forever, and AI can prepare the facts for a judgment call later while you still decide.</p><p>Then rank the repeat pile by hours saved. Multiply the minutes each time by how many times a week you do it. In the sample studio, weekly client updates take 25 minutes for each of 8 clients, which comes to about 3.3 hours a week. The AI mood boards the owner was excited about save about half an hour a week, and they ranked ninth.</p><p>Finally, look at the top of the list and ask the two-minute question. Can I check this result quickly because I know it cold? If the answer is yes, that is your first card.</p><h3>Your memory is a poor accountant</h3><p>One warning about the audit: it usually comes from memory, and memory forgets things. It forgets the eleven small emails, the three reschedules and the thing you do every Thursday without noticing.</p><p>In the Power Lesson that follows the live session, I show the cohort how to let AI read four weeks of their calendar and sent mail, tag the time into the three piles and point out repeat work they forgot to list. It reads and touches nothing, and nothing changes in their ranking until they approve it. The record often turns up a boring, repeated job that never made the handwritten list.</p><h3>Write the card before you open a tool</h3><p>Once you have your task, write a Job Card in five lines before you touch any AI tool.</p><p>The trigger is when the job starts, like Friday at 2 PM for every active project. The task is what exactly gets done, like drafting each client’s weekly update from the project notes. The outcome is what done looks like, like a draft for each client covering what moved, what is next and what we need from them. The check moment is when you read it. The never line is what it must not do, like promise a date or a price, with an instruction to write MISSING when information is missing.</p><p>Notice there are no tool names on the card. A Job Card is a job description for one job, and the tool comes later. Sometimes the right answer is a Claude skill you write yourself, and sometimes a tool you already use does the job well once you set it up carefully.</p><h3>What you get back</h3><p>When you hand off the boring task you understand best, you get more than the hours. You get proof, from your own work, that AI can be trusted with something. That proof is what makes the next handoff easier, and the one after that.</p><p>The pattern I see in my classes is that people who start with the hardest job tend to stall, while people who start with the most ordinary one, check it every week and let their confidence grow tend to keep going.</p><h3>Try this tonight</h3><p>Take ten minutes and write down every task you remember doing this week, with a rough number of minutes next to each. Circle the repeat tasks. For each circled task, ask yourself whether you could check an AI draft of it in two minutes. Put a star next to the ones where the answer is yes, and pick the starred task that saves you the most time.</p><p>That is your first Job Card, and you can write it before bed.</p><p>I wrote the full method up here, with the three piles, the ranking and the five-line card: <em>Which tasks should I hand off to AI first?</em> <a href="https://buildwithaiclub.com/blog/which-tasks-should-i-hand-off-to-ai-first">https://buildwithaiclub.com/blog/which-tasks-should-i-hand-off-to-ai-first</a></p><p>If you want to watch me teach live, my next free AI 101 class at SHACK15 in San Francisco is Wednesday, Nov 18, 6 to 7 PM PT: The One-Person AI Media Setup. Every seat is taken, and the waitlist is open at <a href="https://luma.com/build-kgkg">luma.com/build-kgkg</a>. This week’s Build with AI Club YouTube lesson shows how to turn a card like this into a Claude skill. <a href="https://www.youtube.com/@buildwithaiclub_edu">https://www.youtube.com/@buildwithaiclub_edu</a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=59ef813b3d50" width="1" height="1" alt=""><hr><p><a href="https://medium.com/build-with-ai-club/hand-off-the-boring-task-you-understand-best-59ef813b3d50">Hand off the boring task you understand best</a> was originally published in <a href="https://medium.com/build-with-ai-club">Build With AI Club</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Your customers already asked AI about you]]></title>
            <link>https://medium.com/build-with-ai-club/your-customers-already-asked-ai-about-you-d3aeeaa9fe14?source=rss-f1bc6e9c3b01------2</link>
            <guid isPermaLink="false">https://medium.com/p/d3aeeaa9fe14</guid>
            <category><![CDATA[marketing]]></category>
            <category><![CDATA[chatgpt]]></category>
            <category><![CDATA[small-business]]></category>
            <category><![CDATA[entrepreneurship]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <dc:creator><![CDATA[Arjita Alka Sethi]]></dc:creator>
            <pubDate>Wed, 11 Nov 2026 19:18:58 GMT</pubDate>
            <atom:updated>2026-11-11T19:18:58.974Z</atom:updated>
            <content:encoded><![CDATA[<p>In the AI for Growth cohort last week, I opened the session with one question for the room. Who has ever said “anyone could use this” about their business?</p><p>Hands went up in the room at SHACK15 and reactions came up on Zoom, and if I had been sitting in a chair, my hand would have gone up too. I have said it about every company I have built. It feels like the generous answer, because you don’t want to leave anyone out.</p><p>Then I asked a second question. When someone who needs exactly what you offer opens ChatGPT tonight, what will they type?</p><p>Nobody in the room types “anyone.” And that gap, between how we describe our customers and how our customers describe themselves to AI, is the most useful thing I taught that night.</p><h3>People talk to AI in full sentences</h3><p>For years we searched in fragments. We typed two or three words into a search box and scrolled. Now people open an AI tool and type a whole sentence, and they put themselves inside it. They say who they are, where they live, what is happening and what they are worried about.</p><p>A person no longer searches “bookkeeper.” They ask, “Who is the best bookkeeper for small creative agencies in Oakland?” A person no longer searches “book launch.” They ask, “Where can a debut author hold a book launch in Oakland?”</p><p>Those sentences are a gift to anyone running a business, because the buyer has written your brief for you. They have told you who they are, what they need and the exact words they use for it.</p><h3>Why I teach the question before the persona</h3><p>Personas have helped a lot of businesses, and I have built plenty of them. A persona describes a person: their age, their job, their goals, sometimes their name and a stock photo. The trouble is that we usually write personas from imagination, and imagination is generous to our assumptions.</p><p>A question is different, because somebody actually asked it. It comes with their wording, their worry and their timing already attached. When I put those questions next to the personas people bring to class, the questions almost always tell a sharper story.</p><p>So in Week 2 of AI for Growth, before anyone writes a single page or post, we build a Question Map. You start with one buyer, collect the questions that buyer has already asked, sort them by how close the person is to deciding, and choose the ones you most want to win.</p><h3>Dana picks one buyer</h3><p>I teach with a fictional sample business so nobody’s private data ends up on screen. Her name is Dana, and she runs Juniper Books, an independent bookstore in Oakland.</p><p>Dana could choose many buyers. Book clubs, parents, her regulars and teachers all walk through her door. She picks debut authors in the East Bay who are planning a launch, because a launch night is her highest-value booking and it fills the room with new faces.</p><p>This is the part people resist, so I always say it out loud: Dana still sells books to everyone who walks in. Picking one buyer doesn’t close the door on anyone. It decides who she leads with when she speaks.</p><p>Then we watch her description narrow. “A bookstore” is one of thousands, so AI names the famous ones. “A bookstore in Oakland” is one of dozens. “An Oakland bookstore that hosts author events” is one of a few. “Where debut authors in the East Bay hold their launch night” has only a handful of real answers, and if Dana’s website says it clearly, she can be one of them.</p><h3>The questions are already in your inbox</h3><p>The best news in this lesson is that you don’t have to invent anything. Your buyers have already asked their questions, and they left them in four places.</p><p>They are in reviews, both yours and the ones people leave on businesses like yours. They are in your DMs and emails, the questions people actually sent you. They are in your sales call notes, especially the objections, which are questions in disguise. And they are in FAQs and forums like Reddit, where people ask strangers the things they feel silly asking you.</p><p>When you collect them, copy them word for word. The phrasing is part of the data. “How much does it cost to host a reading?” and “Do indie bookstores charge to host a reading?” are the same question with a different worry underneath.</p><h3>Every question has a mood</h3><p>Once you have a pile, paste it into your AI project and ask it to sort each question by mood. I use three.</p><p>Exploring questions sound like “How do book launches work?” The person is learning, and they need an explainer.</p><p>Comparing questions sound like “What are the best East Bay stores for a reading?” The person has a short list in mind.</p><p>Deciding questions sound like “Is Juniper Books good for a launch?” The person is nearly ready.</p><p>Here is why the moods matter for growth now. Comparing and deciding questions are where AI tools actually name businesses. Exploring questions tend to get a general explanation. So your ten must-win questions come from the comparing and deciding piles.</p><p>Three of Dana’s ten are “Where can a debut author hold a book launch in Oakland?”, “Do indie bookstores charge to host a reading?” and “How do I get 40 people to my first book event?”</p><h3>Look for the gap</h3><p>The last step is my favorite. Look across your questions for the one nobody near you answers well.</p><p>Dana found hers quickly. No bookstore near her explains what a launch night costs or what it includes. Debut authors are asking, the stores are quiet, and AI has nothing clear to quote. That one gap becomes the first page she writes, and later it becomes an article.</p><p>When people build their maps in class, the gap often turns out to be a question about price, process or what happens first, because those are the questions businesses feel awkward answering in public.</p><h3>What this changes for you</h3><p>When you work from real questions, your content stops being a guess. Your website pages answer the questions people actually ask. Your posts use the words your buyers use. And later, when you want to know whether AI recommends you, you already have the ten questions to test.</p><p>It also changes how you feel about marketing. A lot of the founders I teach dread content because it feels like shouting into a room. Answering a real question from a real person feels much more like help, and it is much easier to write.</p><h3>Try this tonight</h3><p>Open your inbox and your DMs and give yourself twenty minutes. Copy ten questions customers or prospects asked you this year, word for word, into one document. Next to each one, write whether the person was exploring, comparing or deciding. Circle the one you most wish AI would answer with your name.</p><p>That circled question is where your growth work starts.</p><p>I wrote up the full step-by-step method here: <em>How do I find the questions my customers are asking AI?</em> <a href="https://buildwithaiclub.com/blog/how-do-i-find-the-questions-my-customers-are-asking-ai">https://buildwithaiclub.com/blog/how-do-i-find-the-questions-my-customers-are-asking-ai</a></p><p>This week’s Build with AI Club YouTube lesson shows the four Claude skills I would set up before anything else, so this kind of work doesn’t have to be explained to AI twice. <a href="https://www.youtube.com/@buildwithaiclub_edu">https://www.youtube.com/@buildwithaiclub_edu</a></p><p>If you’d like to watch me teach live, my free AI 101 class at SHACK15 in San Francisco is tonight, Wednesday, Nov 11, 6 to 7 PM PT. It’s Your AI Research Desk, and free online audit seats are open at <a href="https://luma.com/build-4bhm">luma.com/build-4bhm</a>. Tomorrow, Thursday, Nov 12, 10 to 11 AM PT, Sarah Dooley and I are hosting Aligning AI With Your Family’s Values, free on Zoom, at <a href="https://luma.com/build-nt03">luma.com/build-nt03</a>.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=d3aeeaa9fe14" width="1" height="1" alt=""><hr><p><a href="https://medium.com/build-with-ai-club/your-customers-already-asked-ai-about-you-d3aeeaa9fe14">Your customers already asked AI about you</a> was originally published in <a href="https://medium.com/build-with-ai-club">Build With AI Club</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Stop collecting prompts and start writing skills]]></title>
            <link>https://medium.com/build-with-ai-club/stop-collecting-prompts-and-start-writing-skills-c6de4aef5196?source=rss-f1bc6e9c3b01------2</link>
            <guid isPermaLink="false">https://medium.com/p/c6de4aef5196</guid>
            <category><![CDATA[small-business]]></category>
            <category><![CDATA[prompt-engineering]]></category>
            <category><![CDATA[productivity]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[claude]]></category>
            <dc:creator><![CDATA[Arjita Alka Sethi]]></dc:creator>
            <pubDate>Mon, 09 Nov 2026 19:19:36 GMT</pubDate>
            <atom:updated>2026-11-09T19:19:36.046Z</atom:updated>
            <content:encoded><![CDATA[<p>A few months ago, in a Thursday lab I called Own the System, I put four cards on the screen: context file, skill, project and agent. Then I asked the room a question I had been asking myself for weeks. If you looked at everything you saved in your AI tools this year, how much of it actually gets used?</p><p>The answers were honest and a little embarrassing, mine included. People had folders of prompts copied from posts and courses, documents of “best prompts for marketing,” and notes app pages titled “AI stuff.” Almost nobody opened them. The prompts were good, and they still sat there, because a prompt only helps in the moment you remember to paste it.</p><p>That lab changed how I teach, and it changed how I run my own work. So I want to make a case that sounds small and turns out to be a big shift: stop collecting prompts, and start writing skills.</p><h3>What a prompt actually is</h3><p>A prompt is a request. It is the right tool for a question you will ask once, like “summarize this article” or “give me three subject lines for this email.” You ask, you get the answer and you move on.</p><p>The trouble starts when the same request comes back every week. The Friday update, the meeting prep, the LinkedIn post that has to sound like you, the welcome email to a new client. For those jobs, a prompt turns you into a copy-and-paste machine. You find the prompt, paste it, add the week’s details, notice it forgot your formatting, add the formatting rules, and do the whole thing again next Friday.</p><p>I think of it like hiring someone and then handing them the same sticky note every morning instead of writing them a proper job description.</p><h3>What a skill is</h3><p>A skill is the job description. It is your way of doing one job, written down once and saved to your account. When you describe that situation in a chat, Claude recognizes it and follows your steps, with your examples and your rules.</p><p>Back in March, when I first started building skills for my own business, I described them to Claude as SOPs, and I described the agent that runs them on a schedule as the assistant who follows the SOP. That framing has stuck with me, because every business owner already understands an SOP. You write it once so the work comes out the same way no matter who does it, or how tired you are on a Friday.</p><p>The difference from a prompt is where the knowledge lives. With a prompt, the knowledge lives in your memory and your notes app, and you carry it into every chat. With a skill, the knowledge lives with the AI, and you just say what you need.</p><h3>Context belongs to a body of work, and a method travels</h3><p>The sentence that clears up most of the confusion in my classes is this one. Context belongs to a body of work, and a method travels.</p><p>Your brand rules, your customer and the things you never say are context. They belong inside a project, the workspace for one business, one client or one course. The steps for writing a weekly update are a method. If you use the same steps for three clients, that is one skill, and each client’s details live in their own project.</p><p>When people put the method inside a project, they rebuild it every time they start something new. When they put context inside a skill, the skill gets confused about which client it is talking about. Keeping the two apart is most of the work.</p><h3>My own example: the skill that learns from my edits</h3><p>I write a lot of newsletters across my brands, and for a long time I wrote the same corrections over and over. Write the time from the send date, not the draft date. Use “Build with AI Club” as the name. Make the subject line ask the reader to do something. Cut the punchy one-line endings.</p><p>Now those corrections live in a newsletter skill. It has my voice rules, the layout for each issue and, my favorite part, an edits log. Every time I change a draft, the change gets written down with the rule behind it. When the same correction shows up twice, it moves into the voice rules for good. The skill gets better because I edited it, and I only make each correction once.</p><p>That is the part a prompt library can’t do. A prompt is frozen the day you save it, and a skill grows with you.</p><h3>Why a skill earns its place</h3><p>A skill earns its place when three things are true.</p><p>First, you do the job often enough that retyping the instructions annoys you. If you do it twice a year, write a note instead, because a note is cheaper than a system.</p><p>Second, the job has steps you can describe. If it changes completely every time, that is judgment, and writing it as a procedure makes it worse.</p><p>Third, you have done the job yourself. You can’t write an SOP for a job you have never held, and AI can’t follow one either.</p><p>Those three filters remove a surprising amount of noise. What is left is usually four or five jobs, and those few jobs are where most of the retyping in your week comes from.</p><h3>The part almost everyone gets wrong: the description</h3><p>A skill has a description that tells Claude when to use it. Most people write their steps carefully and then write a lazy description like “helps with emails.” Then the skill never kicks in, and they decide skills don’t work.</p><p>The description is where the effort pays off. Write it in the words you would actually type when you need the job done. “Use this when I paste my messy weekly notes and ask for my update” works far better than “weekly update helper.”</p><p>The test I teach is simple. Close the chat. Open a fresh one and describe the situation without naming the skill. If it doesn’t fire, fix the description, not the steps, because that is nearly always where the problem is.</p><h3>Expect 60 percent, then train it</h3><p>The first time you run a new skill, expect it to get about 60 percent right. That is normal. The part people skip is what comes next. When the draft is close but wrong, most people fix the draft and move on, and the same mistake comes back next week.</p><p>Fix the skill instead. Ask Claude to tell you the exact line to add to the skill so the mistake doesn’t happen again, read it and save it. Your corrections are what take it from 60 percent toward 90. This is the middle step of the framework I teach, learn AI, train AI and implement AI, and training is the step most people skip.</p><h3>Where agents come in</h3><p>Once a skill works reliably, you can put it on a schedule and let it run as an agent, and that is when it starts to feel like having help. The agents I teach follow the same four steps every time: pull from where the work happened, draft it in your voice, stop and show you, and send only after you approve.</p><p>I keep that third step on purpose. Fully automated sending is possible, and I still want the last connection with a person to be made by a person. AI removes the chore, and the relationship stays yours.</p><p>The agent is only as good as the skill underneath it, though, which is why skills come first.</p><h3>What changes when you stop collecting</h3><p>When I stopped collecting prompts, my AI setup got smaller and my results got better. I have fewer things saved and I use all of them. I no longer go looking for the perfect prompt, because the work I repeat is already written down the way I like it.</p><p>The bigger change is in how I think about AI. A prompt library treats AI like a vending machine, where you find the right button and hope for the right snack. A skill treats AI like your first hire, someone you train once, correct often and trust more over time. You lead AI, and the skill is how you write down the way you lead.</p><h3>Try this tonight</h3><p>Pick one job you explained to AI more than once last week. Open a blank document and write it the way you would brief a new assistant: the steps, what a good result looks like, one example you were happy with and one line it must never cross. Then write one sentence that says when this job comes up, in your own words. That sentence is your description, and the rest is your first skill.</p><p>Tomorrow, save it in Claude and test it in a fresh chat without naming it.</p><p>If you want the step-by-step version, I wrote it up here: <em>What is a Claude skill, and how is it different from a project?</em> <a href="https://buildwithaiclub.com/blog/what-is-a-claude-skill-and-how-is-it-different-from-a-project">https://buildwithaiclub.com/blog/what-is-a-claude-skill-and-how-is-it-different-from-a-project</a></p><p>This week’s Build with AI Club YouTube lesson walks through the four skills I would set up before anything else. <a href="https://www.youtube.com/@buildwithaiclub_edu">https://www.youtube.com/@buildwithaiclub_edu</a> And if you want to watch me teach live, my free AI 101 class at SHACK15 in San Francisco is Wednesday, Nov 11, 6 to 7 PM PT, with free online audit seats at <a href="https://luma.com/build-4bhm">luma.com/build-4bhm</a>. On Thursday, Nov 12, 10 to 11 AM PT, Sarah Dooley and I are hosting Aligning AI With Your Family’s Values, free on Zoom, at <a href="https://luma.com/build-nt03">luma.com/build-nt03</a>.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=c6de4aef5196" width="1" height="1" alt=""><hr><p><a href="https://medium.com/build-with-ai-club/stop-collecting-prompts-and-start-writing-skills-c6de4aef5196">Stop collecting prompts and start writing skills</a> was originally published in <a href="https://medium.com/build-with-ai-club">Build With AI Club</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Why I make my students brainstorm with the laptop closed]]></title>
            <link>https://medium.com/@arjita_sethi/why-i-make-my-students-brainstorm-with-the-laptop-closed-1056cc4a6c63?source=rss-f1bc6e9c3b01------2</link>
            <guid isPermaLink="false">https://medium.com/p/1056cc4a6c63</guid>
            <category><![CDATA[startup]]></category>
            <category><![CDATA[education]]></category>
            <category><![CDATA[entrepreneurship]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[founders]]></category>
            <dc:creator><![CDATA[Arjita Alka Sethi]]></dc:creator>
            <pubDate>Wed, 04 Nov 2026 21:11:31 GMT</pubDate>
            <atom:updated>2026-11-04T21:11:31.573Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*Pqm51eddmMtmeeQhjSiFAw.jpeg" /></figure><p>A few weeks ago, my students came back from a breakout room, and before I could say anything, one of them unmuted and asked, “Just to confirm, we were not supposed to use AI in that round, correct?”</p><p>He was right. For ten minutes, each team had to come up with three different solutions to the problem they had found in their customer interviews, and they had to do it without any AI tool. After that round, I opened Replit and Claude and showed them how to build all three solutions in about an hour. The building was the fun part, and I wanted them to earn it.</p><p>I teach entrepreneurship at SF State, and I run AI education for non-technical founders, so it can sound strange that I ask people to close the laptop. I use AI every day. I use Claude for so much of my life that when I asked it to mock up a sample business in class, it started suggesting layouts based on my own companies. What I ask is narrower than general caution: I want people to choose carefully which part of the work they hand to it.</p><p>Here is the comparison I use. Think about cooking for your family. A good kitchen gadget can chop the onions, knead the dough and keep the rice from burning, and it saves you an enormous amount of time. It cannot tell you what your family is hungry for tonight, because it has never sat at your table. You know that, because you were at the table when someone pushed the plate away last week.</p><p>AI tools were trained on what is already public on the internet. When you ask one what business you should start, it gives you a thoughtful, well-written version of the average answer, built from millions of ideas that already exist. That is a useful thing to have in many situations. For a founder, the average answer is the one place you cannot afford to start, because the average idea already has a hundred competitors.</p><p>Your edge lives somewhere AI has never been: the interview where a customer told you what she already tried and why she gave up, along with your notes, your own story and the specific people you know. When a team sits together with pen and paper and those notes, the three solutions they come up with are strange and specific and sometimes a little uncomfortable, and that is exactly what I want to see.</p><p>In class I gave an example. If the problem is that people in tech in San Francisco sit all day and barely move, one solution is a social fitness app, another is an office running group that meets every Monday at 5 PM, and a third is a website listing affordable active events around the city. I came up with those by thinking about real people I know. An AI tool might have offered one of them, and it probably would not have offered all three as a set, because the set came from noticing what different people actually do after work.</p><p>Once the thinking is done, I want people to use AI fully and without guilt. Build all three prototypes. Make them fast and a little ugly. Let Claude ask you clarifying questions, let Replit publish a live version with a free domain, and then take all three back to the same people you interviewed and let them choose. This is the part of entrepreneurship that used to cost months and thousands of dollars, and now a non-technical founder can do it in an afternoon. I find that genuinely moving, because so many of my former students had good ideas and no way to test them.</p><p>I want to be generous here, because everyone is in their own season with AI. If you have been asking ChatGPT what business to start, you are not doing anything wrong. You are curious and you are moving, and that already puts you ahead of where you were. I simply want you to notice what came back, and to ask whether it sounds like you or like everyone. If it sounds like everyone, go back to your notes.</p><p>The way I say it in every session is that you lead AI, and AI does not lead you. In practice, leading looks very ordinary, because it usually means ten quiet minutes with a notebook before you open any tool and a decision to trust what your customer told you over what the internet averages out to.</p><p>So before you open a single tool this week, close the laptop, write down what your customers actually said, and come up with three answers that only you could have written.</p><p>If you want a simple path from that notebook page to a live first version, I put the steps into a free one-page “Launch your AI business” checklist from New Founder School. You can find it through the link in my profile.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=1056cc4a6c63" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Nobody in the room is more than eight weeks ahead of you]]></title>
            <link>https://medium.com/build-with-ai-club/nobody-in-the-room-is-more-than-eight-weeks-ahead-of-you-fe194e749e27?source=rss-f1bc6e9c3b01------2</link>
            <guid isPermaLink="false">https://medium.com/p/fe194e749e27</guid>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[education]]></category>
            <category><![CDATA[entrepreneurship]]></category>
            <category><![CDATA[productivity]]></category>
            <category><![CDATA[machine-learning]]></category>
            <dc:creator><![CDATA[Arjita Alka Sethi]]></dc:creator>
            <pubDate>Wed, 04 Nov 2026 20:54:56 GMT</pubDate>
            <atom:updated>2026-11-04T20:55:36.129Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*9mHsCLHfdWlGGnYMgapsSQ.jpeg" /></figure><p>A few weeks ago I started a virtual workshop the way I start every session. I asked each person three questions: who are you, what industry are you in, and how comfortable are you with AI on a scale of one to ten.</p><p>The answers covered the whole scale. One woman, an account manager at a steel fabrication company, said 1.5. A marketing strategist who runs her own agency said she was an 8 or a 9. In between there was a real estate investor, a brand new real estate agent who had just returned to work after ten years at home with her kids, a coach, a notary building a product, and a sales executive planning travel across four cities in one week.</p><p>Then something interesting happened. As people described what frustrated them, the 1.5 and the 9 said almost the same thing. They were prompting back and forth for ages and still not getting what they needed. One person put it perfectly: she didn’t want a two-hour conversation with AI for a task she could do herself in half the time.</p><p>I have been thinking about that room ever since, because it showed me something I believe about this moment. The gap between the person who feels hopelessly behind and the person who feels fluent is much smaller than either of them thinks.</p><p>Here’s why. I told that room that if I had taught the same session eight weeks earlier, it would have been a completely different session, and if I teach it eight weeks from now, it will be different again. I say this in almost every class, because it keeps being true. At SHACK15 in September, I opened a session on AI agents by saying the training I had given eight weeks before was already out of date. The tools had changed that much.</p><p>When the ground moves that fast, nobody gets to stand still at the front. The most advanced person in any room I teach is usually four, six, maybe eight weeks ahead of the newest person. That is the whole lead. I live in San Francisco, I build with these tools every day, and I still think most of us use our AI tools at something like ten percent of what they can do. Me included.</p><p>I find that freeing, and I want you to find it freeing too. If you are at a 1.5 right now, you aren’t years behind. You’re a few weeks of practice behind, and those weeks are available to you starting tonight.</p><p>There’s a quieter point underneath this. The 1.5 and the 9 had the same problem because the problem was never skill. Neither of them had trained their AI. It didn’t know who they were, what they ran or how they liked their answers, so it gave both of them the average answer it gives everyone. Being a power user didn’t protect the marketing strategist from that, and being new didn’t doom the account manager to it. The fix was the same for both of them, and it took about thirty minutes.</p><p>So I would gently push back on the way we talk about AI fluency, as if it were a ladder some people are already at the top of. I see it more like a walk everyone started at roughly the same time, where some people have been walking a little longer. The person ahead of you can tell you where the puddles are. They can’t walk it for you, and they aren’t much farther down the road.</p><p>This is also why I don’t want anyone chasing every new tool to catch up. When everything changes every eight weeks, the tool you rush to learn today may be gone or rebuilt by spring. I would much rather you pick one of the established tools, like Claude, ChatGPT or Copilot, train it on your life and keep showing up. In a season like this one, the people who keep practicing are the ones who stay a few weeks ahead.</p><p>And it’s why every session I teach starts with paper and pen, before anyone opens a laptop. I ask people to write down the one task they wish would disappear from their week. The account manager and the agency owner both wrote something down. Neither needed any AI skill to do it. That page is where training starts, with you knowing what you want before you ask a machine for help.</p><p>Everyone in that room is in their own season. Some are learning AI for the first time, and some are teaching it to their clients. From where I stand, both of them are early, and I’m glad they started.</p><p>If you are reading this on a Wednesday, I’m teaching tonight. Bring a notebook, bring whatever you’re using right now, and we’ll start from exactly where you are.</p><p>The free Build with AI Club Wednesday session is tonight, November 4, 6 to 7 PM PT, at SHACK15 in San Francisco or online.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=fe194e749e27" width="1" height="1" alt=""><hr><p><a href="https://medium.com/build-with-ai-club/nobody-in-the-room-is-more-than-eight-weeks-ahead-of-you-fe194e749e27">Nobody in the room is more than eight weeks ahead of you</a> was originally published in <a href="https://medium.com/build-with-ai-club">Build With AI Club</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[My hands get cold first]]></title>
            <link>https://medium.com/@arjita_sethi/my-hands-get-cold-first-3bb0f72b01dc?source=rss-f1bc6e9c3b01------2</link>
            <guid isPermaLink="false">https://medium.com/p/3bb0f72b01dc</guid>
            <category><![CDATA[mindfulness]]></category>
            <category><![CDATA[ayurveda]]></category>
            <category><![CDATA[wellness]]></category>
            <category><![CDATA[mental-health]]></category>
            <category><![CDATA[self-care]]></category>
            <dc:creator><![CDATA[Arjita Alka Sethi]]></dc:creator>
            <pubDate>Mon, 02 Nov 2026 19:55:47 GMT</pubDate>
            <atom:updated>2026-11-02T19:55:47.594Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*kGstUSDBbP1om8QHJJX-pQ.jpeg" /></figure><p>Every year around this time, I am the first person in the room to reach for a shawl. My hands go cold before anyone else’s, my skin starts asking for more oil, and my mind starts running a lap ahead of my day. For a long time I filed all of that under one word, which was busy.</p><p>I am a founder and a mother, and busy is an easy word to hide behind. Light sleep meant I had a lot on my plate. Starting five things and finishing two meant the same, and so did forgetting lunch and eating crackers at the counter at four. I treated every one of these as a small personal failure that I could fix with more discipline next week.</p><p>Then I learned the Ayurvedic name for this time of year, and it changed the way I talk to myself.</p><p>In Ayurveda, late fall into early winter is Vata season. Vata is the dosha made of air and space, and its qualities are cold, dry, light and moving. When the weather outside turns cool, dry and windy, those same qualities rise inside us. My dosha type is Vata-Pitta, Vata dominant, so I feel it first and I feel it hard. The cold hands, the dry skin, the 3 a.m. wake-ups and the scattered mind were never a character flaw. They were the season showing up in a body that is built to notice it.</p><p>I want to say something gently here, because I think a lot of us carry the same habit. When we feel off, we look for what we did wrong. We assume we need a stricter plan, an earlier alarm or a better app. Some of the time that is true. A lot of the time, though, the world around us changed and we kept living as though it hadn’t. We are still eating cold salads at our desks in November. We are still drinking iced coffee while the wind rattles the windows. Our bodies are responding to the weather exactly the way a body should.</p><p>Ayurveda gave me a kinder first question. Before I ask what is wrong with me, I ask what season it is. That one question takes a lot of shame out of the room, because it turns a personal problem into a natural pattern, and natural patterns have natural responses.</p><p>The response to Vata is warmth and rhythm, which in practice means warm, cooked food with a little ghee, warm oil on the skin before a shower, and lunch at the same time each day with a bedtime before the second wind arrives. None of this is dramatic. We do need to be told, though, because modern life has trained most of us to override the season instead of listening to it.</p><p>I also want to be honest that I do not do this perfectly. Some weeks I eat well and oil my skin every morning, and some weeks I am on my laptop at eleven at night with cold feet and a cup of something I should not be drinking. Ayurveda does not ask me to be flawless. It asks a gentler question, which is whether I can do a little better from wherever I am today. In a hard week, that might be one warm meal. In an easier week, it might be the whole morning routine.</p><p>What I love about this way of thinking is that it works for everyone, whatever their dosha type and whatever their season of life. A Pitta type will feel autumn differently from me, and a Kapha type differently again, and every one of us is allowed to respond to what our own body is saying. I find that very freeing as a teacher, because I never have to hand anyone a rulebook. I get to hand them a lens.</p><p>So if you are reading this with cold hands and a mind that won’t settle, I would love for you to try my question this week. Before you decide something is wrong with you, ask what season it is, and then ask what your body might need from the season.</p><p>If you want to know how your own body moves through Vata season, the free dosha quiz on findshaanti.com is a good place to begin.</p><p>So this week, check the weather before you decide anything is wrong with you, and let me know what you notice.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=3bb0f72b01dc" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Revenue is the best money]]></title>
            <link>https://medium.com/@arjita_sethi/revenue-is-the-best-money-38ff87190a1b?source=rss-f1bc6e9c3b01------2</link>
            <guid isPermaLink="false">https://medium.com/p/38ff87190a1b</guid>
            <category><![CDATA[startup]]></category>
            <category><![CDATA[small-business]]></category>
            <category><![CDATA[founders]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[entrepreneurship]]></category>
            <dc:creator><![CDATA[Arjita Alka Sethi]]></dc:creator>
            <pubDate>Mon, 02 Nov 2026 19:39:09 GMT</pubDate>
            <atom:updated>2026-11-02T19:39:09.032Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*M04hoEtHnSNcFlBprrqmOg.jpeg" /></figure><p>Every semester I teach a session on how founders pay for their first business. This semester I moved it from a recorded lecture to a live class, because the rules changed fast enough that I did not want to teach from last year’s slides.</p><p>Here is what changed. AI made building cheap. A few years ago, most people who wanted to start something believed the first step was raising money, because building a website, an app or a product cost so much before a single customer saw it. That is no longer true. You can now describe what you want in plain words and have a working first version the same week. I built Shaanti, my Ayurvedic wellness platform, in 48 hours this way.</p><p>So the question I asked my students was simple. If you can build it, why are you waiting for someone’s permission to start?</p><p>Living inside the San Francisco bubble, it is easy to believe that raising money is the definition of success. I see it differently. The best money a business can receive is revenue, because revenue is a customer telling you that what you built is worth paying for. A business that makes money gives you choices. It lets you decide how fast to grow, how much to work, and who you build it with.</p><p>The other shift I find exciting is who is starting companies. I shared a Stripe Atlas figure with my students: in Q2 2026, 63 percent of companies incorporating on the platform had a single founder. One person with AI can now go much further on their own than a small team could a few years ago.</p><p>What should a first-time founder actually do? I gave my students three directions, and I would give you the same ones.</p><p>Sell before you build. A waitlist, a presale or a first paid month tells you more than any plan. Customers who pay early fund the thing they want and prove demand at the same time.</p><p>Build small and fast. Your first version needs a page, a form, a payment link and a way to follow up. AI can help you build all of that in an afternoon.</p><p>Count real customers. Instead of dreaming about a share of a huge market, write down how many people you can reach, how long they will stay and what they will pay. It is humbler math, and it tells you whether the business can pay you.</p><p>When I left that class, the line I kept coming back to was this: you do not need anyone’s permission to start a business anymore. The customers you win first are what give you every option after that.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=38ff87190a1b" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Learn AI, train AI, implement AI: the step everyone skips]]></title>
            <link>https://medium.com/build-with-ai-club/learn-ai-train-ai-implement-ai-the-step-everyone-skips-06bfe95f14ca?source=rss-f1bc6e9c3b01------2</link>
            <guid isPermaLink="false">https://medium.com/p/06bfe95f14ca</guid>
            <category><![CDATA[productivity]]></category>
            <category><![CDATA[entrepreneurship]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[future-of-work]]></category>
            <category><![CDATA[leadership]]></category>
            <dc:creator><![CDATA[Arjita Alka Sethi]]></dc:creator>
            <pubDate>Mon, 02 Nov 2026 19:23:24 GMT</pubDate>
            <atom:updated>2026-11-02T22:13:44.311Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*4WqJjM1jtDOlDHL_JTVxGg.jpeg" /></figure><p>When someone tells me AI doesn’t work for them, I usually ask them what they did right before they gave up. The answer is almost always the same. They tried a few prompts, got generic answers, and decided the tool was overhyped.</p><p>I think about it the way I think about hiring. If you hired a brilliant new assistant and never told them who you are, what you run, or how you like things done, you would get polite, generic work for a long time. Nobody would blame the assistant for that. With AI, we tend to blame the tool.</p><p>The framework I teach is three steps: learn AI, train AI, implement AI. Most people do the first step and jump to the third. They learn what the tools can do, and then they try to put them to work right away. The middle step, training, is the one that makes everything else work, and it is the one almost everyone skips.</p><p>Training means giving the AI the context it needs to work like it knows you. In my workshops, we build what I call a brain project. It holds an instructions card that says who you are, what you run, what you are working on right now, and how you like your answers. It holds your real documents, in formats the AI reads well. When the output is bad, you fix it by improving the instructions or by asking the AI what context it is missing, instead of re-prompting the same question over and over.</p><p>I learned this the humble way. I have been building AI projects for years and I teach project setup publicly, so I thought my setup was complete. My business context was all there. What was missing was my life: my household, my energy, my real calendar. The answers I got were good for the business and blind to everything around it. Once I added that context, the AI started giving me plans I could actually live with.</p><p>The second reason people skip training is speed. Every week there is a new tool, a new feature, a new demo. It feels like falling behind if you are not chasing all of it. My advice is the opposite. Wait for tools to stabilize before you rebuild your workflow around them, and stay consistently present instead, because AI changes daily and the people who do well are the ones who keep showing up and adjusting.</p><p>This is also why I keep saying the same sentence in every session: you lead AI, AI does not lead you. When you skip training, you let the tool decide what good looks like, and it will pick the average. When you train it, you decide. Your judgment, your taste and your context become the standard the AI works to.</p><p>If you want to try this today, start small. Write a one-page instructions card about yourself and your work. Add two or three documents that show how you actually think. Then ask the AI the same question you asked last week and compare the answers. The difference is usually obvious within a few minutes, and from there, implementing AI in real work stops feeling like a gamble.</p><p>I walk through the full build, step by step, in this week’s lesson on YouTube: <a href="https://youtu.be/fTPYGa6Lq1c">https://youtu.be/fTPYGa6Lq1c</a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=06bfe95f14ca" width="1" height="1" alt=""><hr><p><a href="https://medium.com/build-with-ai-club/learn-ai-train-ai-implement-ai-the-step-everyone-skips-06bfe95f14ca">Learn AI, train AI, implement AI: the step everyone skips</a> was originally published in <a href="https://medium.com/build-with-ai-club">Build With AI Club</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
    </channel>
</rss>