<?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 Small Teams Wrangling AI on Medium]]></title>
        <description><![CDATA[Stories by Small Teams Wrangling AI on Medium]]></description>
        <link>https://medium.com/@teambrain?source=rss-6854dff92520------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*dO6RS0iz6AlcTTyAil7aug.png</url>
            <title>Stories by Small Teams Wrangling AI on Medium</title>
            <link>https://medium.com/@teambrain?source=rss-6854dff92520------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Wed, 18 Nov 2026 04:59:08 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@teambrain/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[Five people on your team are explaining a different company to AI]]></title>
            <link>https://medium.com/@teambrain/five-people-on-your-team-are-explaining-a-different-company-to-ai-03593e13e82a?source=rss-6854dff92520------2</link>
            <guid isPermaLink="false">https://medium.com/p/03593e13e82a</guid>
            <category><![CDATA[claude]]></category>
            <category><![CDATA[chatgpt]]></category>
            <category><![CDATA[small-business]]></category>
            <category><![CDATA[productivity]]></category>
            <category><![CDATA[ai]]></category>
            <dc:creator><![CDATA[Small Teams Wrangling AI]]></dc:creator>
            <pubDate>Fri, 13 Nov 2026 12:01:02 GMT</pubDate>
            <atom:updated>2026-11-13T12:01:02.300Z</atom:updated>
            <content:encoded><![CDATA[<p>Everyone keeps their own pasted blurb about tone and customers. Each copy drifts, and the new hire has no copy at all.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*5ZMo_PKyrl4wuJS07IZgBA.png" /></figure><p>Every morning you open a chat and paste in the paragraph about what your company does, who your customers are and how you like things written. Only then do you ask for the thing you actually wanted.</p><p>That paragraph is a copy. Somewhere there’s the real version of your company, the one that changes when you raise a price, lose a client or decide the newsletter goes out every two weeks. What you paste is a snapshot of it from whenever you last updated the note. Every workaround people build for re-explaining context to AI makes that copy easier to carry, and none of them deals with the fact that it’s a copy.</p><h3>Why the copy exists at all</h3><p>A new chat starts empty. The AI only knows what is in that conversation, so if your business context is not in it, the AI guesses. The tool has no place of its own to keep what you’ve told it, so the context has to arrive from somewhere every single time.</p><p>Right now that somewhere is you. Your memory, your notes file, your clipboard. You’re the one who brings the copy to each chat, and that’s a job you’ve been doing by hand every day, which is why willpower alone won’t make it go away.</p><h3>The master blob</h3><p>The obvious fix is a master context doc. Write the company explainer once, keep it in a note, paste it at the top of every chat.</p><p>It beats retyping, and for a while it works. Give it a few months, though. The blob goes stale, because you update the business and forget to update the note. It grows until you’re pasting three pages to ask a two-line question. And it only covers the general stuff. What a real task needs, like what you decided about that client last week, isn’t in it, so you type that part fresh anyway.</p><p>That’s the copy problem in its simplest form. The company moved on, the note didn’t, and nothing tells you which lines have gone out of date.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*QwRIzFurak9oDXbH8eJwLw.png" /></figure><h3>Custom instructions and memory are copies too</h3><p>The tools offer partial fixes, and they’re worth using for what they are.</p><p>Custom instructions save standing context in your account so every chat starts with it. Good for the stable stuff, like what the company does and how you like answers formatted. But it’s a copy kept in a small box. There’s a size limit, so it can’t hold a knowledge base, and it’s one more place to remember to update when something changes.</p><p>Memory features go further. As of 2026, ChatGPT can carry things across your chats and Claude can look back at your past conversations. That is genuinely useful, and if you work alone it closes a lot of the gap. But look at where it copies from: your chats, on your account. It learns the version of the company that came up in your conversations. It doesn’t hold your project details, your process docs or the decisions your team made in a meeting you weren’t in, and none of it reaches anyone else at your company.</p><p>So each of these is a better way to keep a personal copy, and on a team, personal copies are where the trouble starts.</p><h3>Five copies of one company</h3><p>Alone, re-explaining costs you a few minutes a day. On a team, everyone makes their own copy.</p><p>Your version of the company blurb says one thing about tone. Your ops person’s says something slightly different, because she wrote hers months ago and you’ve rewritten yours since. The new hire has no copy at all, so her AI drafts sound like a stranger wrote them. Five people are each doing the job by hand, and the five copies have drifted apart. Each AI works from what it’s handed, and what it’s being handed is five slightly different companies.</p><p>Picture the week you change your pricing. You update the website and the sales sheet, and you mean to tell everyone. Your own pasted blurb gets the new number, because you’re the one who changed it. Your ops person’s doesn’t, because her note sits in her own files and nobody thought to touch it. A few weeks later a customer replies to a quote she drafted with her AI, asking why it doesn’t match the website. Nobody was careless. The copy she was carrying was simply older than yours.</p><p>That’s also why the drift is easy to miss. Each person sees their own AI doing fine. It only shows up where the copies meet, in front of a customer or in a draft someone else has to fix.</p><h3>What the new hire inherits</h3><p>The new person is where the copy problem is plainest. On her first day she has nothing to paste, so she does the sensible thing and asks around. Someone forwards their blurb. Someone else forwards a different one. She now has two copies of the company and no way to tell which is current, so she picks one and starts building a third in her own chat history.</p><p>At the end of our VA onboarding, we connect the VA to the same AI tools the rest of us use, so they aren’t working outside our systems. I think that’s the step most onboarding leaves out, and it decides whether a new person’s AI knows the company on day one or spends the first few weeks guessing.</p><h3>Stop making copies</h3><p>Here’s the simpler way to think about it: stop carrying copies to the AI, and give it the original to read from instead.</p><p>That means one shared workspace that holds what your company knows: what you sell, who your customers are, how you write, and what has been decided. You write it down once, keep it current in that one spot, and connect your AI to it. Every chat then starts with the AI already reading from that source. Nobody pastes a blob, and nobody re-explains the client situation, because it’s written down where the AI can see it.</p><p>Our founder’s working day runs on this. Most mornings start with talking to Claude, out loud rather than typing, across three VS Code windows for three different projects, plus Claude Desktop for social content. When a day goes into marketing strategy, that work gets saved to the shared workspace. When two hours go into the brand voice doc, that gets saved there too. Because the AI gets the same context every time, a VA could drive it directly off that same saved work and produce the same kind of output.</p><p>This is what TeamBrain is for. Your team’s context lives in one workspace, you connect Claude or ChatGPT to it, and the re-explaining goes away, because the context is no longer something a person has to deliver.</p><p>Everyone’s AI reads the same source, so the drift stops too. There’s one version of the tone and one record of each decision, and the new hire’s AI knows as much as yours on her first day.</p><p>Picture that pricing week again with one original. You change the number in the workspace, once. The next quote your ops person drafts comes back with the new price, because her AI read it from the same place yours did, and nobody had to remember to update a note.</p><h3>What to keep and what to move</h3><p>You don’t have to throw out what you’ve built. Custom instructions are still a fine place for personal preferences, like how you like answers formatted. Memory still helps you pick up where you left off in your own work. What moves is the company itself: the pricing, the tone, the decisions, the client history. That belongs in the one place everyone’s AI reads.</p><p>A good first step is to take the blob you already paste, put it into the shared workspace as the starting version, and connect one AI to it. Then add what the blob never had, the decisions from the last few weeks, and notice how much less you type before each request.</p><p>The test is simple. Open a fresh chat, ask for something real without pasting anything first, and see if the answer sounds like your company. When it does, you’re done re-explaining.</p><p><em>If you’ve built a context doc or custom instructions for this and it broke down somewhere, I’d like to hear where. Responses are open.</em></p><p><em>I’m Maya. I write for TeamBrain about getting a whole small team working with AI, not just one person, and everything here comes from how our own team works, from the project manager to the VAs to the clients. If you want the bigger picture, start with our free guide,</em></p><p><a href="https://teambrain.co/free-guide/?utm_source=medium&amp;utm_medium=article&amp;utm_campaign=stop-re-explaining-context-to-ai"><em>The New Way of Working with AI</em></a><em>.</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=03593e13e82a" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Most small business owners skip the one step that makes AI useful]]></title>
            <link>https://medium.com/@teambrain/most-small-business-owners-skip-the-one-step-that-makes-ai-useful-98a9545f15bd?source=rss-6854dff92520------2</link>
            <guid isPermaLink="false">https://medium.com/p/98a9545f15bd</guid>
            <category><![CDATA[chatgpt]]></category>
            <category><![CDATA[claude]]></category>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[productivity]]></category>
            <category><![CDATA[workflow]]></category>
            <dc:creator><![CDATA[Small Teams Wrangling AI]]></dc:creator>
            <pubDate>Fri, 06 Nov 2026 12:01:02 GMT</pubDate>
            <atom:updated>2026-11-06T12:01:02.441Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*2mo2Vg22nCt3IpdzQR0Oqw.png" /></figure><p>You run a pizza shop, a small nonprofit or a two-person place that fixes bikes, and everyone keeps telling you to “use AI.” Most of the advice assumes you already know the jargon, and the one step that decides whether your first try works is the one beginners rush past.</p><h3>Picture a new hire on her first Monday</h3><p>Say you take on a new hire. She’s sharp, she types fast and she wants to help, but she has never set foot in your shop. If you hand her a notepad on her first morning and say “write this week’s specials post,” you’ll get something back. It will be friendly and tidy, and it could be about any pizza place in the country, because she doesn’t know yet that your thing is big New York slices, that most of your customers are students, or that you’d never in your life call a slice “artisanal.”</p><p>Nobody actually treats a new hire like that. You’d spend ten minutes at the counter with her first: who comes in, what you’re known for, how you talk to people, what this week’s deal is. Then you’d ask for the post, and it would come back sounding like your shop.</p><p>A free AI chat tool starts every job where she started that Monday morning. It’s software, and it knows nothing about your shop until you tell it. Most beginners skip the ten minutes at the counter, get the generic post back, and decide AI isn’t for them. Everything below is about not skipping it.</p><h3>Most small businesses haven’t started yet either</h3><p>The way AI gets sold, you’d think every other business has it figured out and you’re the last one to show up. Up close it doesn’t look like that. Most small businesses are right where you are: curious, a little unsure, and yet to do the first thing.</p><p>That makes you early, which is a good place to be. Do one small, useful thing now and you’ll be comfortable with it long before it matters to your bottom line.</p><h3>Pick the boring writing job you do every week</h3><p>Skip “transform the business” for now. The first step is small on purpose, because a small win you actually finish beats a big plan you never start.</p><p>Think about the writing task you do over and over and quietly dread. Every business has one:</p><ul><li><strong>Pizza shop:</strong> the weekly specials post for Facebook and Instagram.</li><li><strong>Nonprofit:</strong> the thank-you email to everyone who donated.</li><li><strong>Bike shop:</strong> the reply you type ten times a week to “do you have this in stock, and how much.”</li><li><strong>Cleaning service:</strong> the quote email you rewrite from scratch for every new customer.</li></ul><p>Pick the one that makes you sigh a little when it lands on your plate. That’s the first job you hand over, the same way you’d start a new hire on one task before you show her the whole shop.</p><h3>Your first step takes about fifteen minutes</h3><h3>1. Open a free AI chat tool</h3><p>Go to Claude or ChatGPT in your browser and make a free account. There’s nothing to install and no credit card for the free version. It’s a text box you type into, a bit like texting someone who’s fast and never gets tired.</p><h3>2. Give it the first-day briefing</h3><p>This is the ten minutes at the counter, and it’s the step most beginners skip. Instead of just asking for “a social post,” tell it what you’d tell the new hire: who you are, who you’re talking to, what the offer is and how you sound. In your own words, something like this:</p><blockquote><em>I run a small pizza shop in a college town. Our thing is New York style, big slices, cheap. This week’s special is two slices and a soda for six dollars, weekdays until 4pm. Write me a short, friendly Facebook post about it. Keep it casual, no corporate voice.</em></blockquote><p>That’s five sentences, and it covers everything the new hire would have needed. You get back a post in your voice instead of a stranger’s.</p><p>We learned this with people long before we learned it with AI. When a new VA joins our team, we record the onboarding call instead of handing over a written manual, because written instructions don’t stick and some people only really take things in by watching. Every task they get after that comes with a clearly defined deliverable, so they know what finished looks like. A VA with the full background and a clear ask does good work in her first week, and one without it spends that week guessing. The chat box runs on the same rule. The background and a clear ask are most of the result.</p><p>One small habit makes this easier from week two. Keep your briefing paragraph in a note on your phone or computer, and paste it in at the start of each new chat instead of writing it from scratch. Update the deal each week and leave the rest alone.</p><h3>3. Talk back to it</h3><p>The first answer is a draft. This is the part people don’t realize they’re allowed to do: you give it notes, the way you would with the new hire.</p><ul><li>“Shorter.”</li><li>“Make it funnier.”</li><li>“Add that we deliver.”</li><li>“That sounds too salesy, tone it down.”</li></ul><p>Go back and forth two or three times and in a couple of minutes you’ve got a post you’d actually publish. You say what you want and it does the typing. You don’t even have to type your side. Our founder does most of this out loud, talking rather than typing, and the microphone button on your phone’s keyboard does the job fine for a specials post.</p><h3>4. Read it, fix it, use it</h3><p>You’re still the boss. You’d read a new hire’s first post before it went up, and this is no different. Change the one word that isn’t quite you. Fix anything that’s off, because it can get facts wrong and you know your business better than it does. Then post it.</p><p>That’s the whole loop: pick a task, give it the briefing, talk back, check the result. You just used AI in your business.</p><h3>Run the same loop every week</h3><p>Run that loop on the same task every week and it gets faster each time. The briefing is already written, you know which notes it usually needs, and soon the job you dreaded takes five minutes instead of forty.</p><p>Then, and only then, pick a second task. Maybe the donor thank-you email, maybe the stock-and-price reply. Add them one at a time, the way you’d give a new hire more of the shop once she has the first job down. Bit by bit you hand off the boring writing and keep the parts only you can do.</p><p>A few honest notes before you start:</p><ul><li><strong>Check the facts.</strong> It sounds confident even when it’s wrong. Prices, dates, names and hours get your eyes every time.</li><li><strong>Keep private things private.</strong> Customer card numbers, home addresses and anything sensitive stay out of the chat box.</li><li><strong>Your voice is the point.</strong> You’re using it to get your own voice out the door faster, so keep tuning the briefing until the drafts sound like you.</li></ul><h3>When there are a few of you, the briefing has to be shared</h3><p>For now, one person and one chat window is plenty. That’s the right size for where you are.</p><p>Picture it a few months on, though. You’ve got the specials post down. Your manager uses the same kind of tool for the staff schedule note, and your part-timer uses it to answer reviews. Each of them wrote their own briefing, in their own words, in their own account, and each one describes a slightly different shop. One still has last month’s deal in it. Another never mentions that you deliver.</p><p>That’s the new hire problem again, three times over, and everyone is re-explaining the business from scratch every time they open a chat. At that point you’ll want the whole team working from the same background. It’s why the last step of our own VA onboarding is connecting the new person to the same AI tools the rest of us use, so nobody is working from their own private version of the business. There are tools for that when you get there, TeamBrain among them, a shared brain your team and their AI can both work from. As of 2026 the free plans from Claude and ChatGPT are built around one person in one chat, so sharing that context across a team is the thing you’ll go looking for.</p><p>But that’s later. This week the step is simpler. Open the chat, pick the one task you dread, and give it the briefing you’d give a new hire on her first Monday. See what comes back, and after a week it stops feeling strange.</p><p><em>If you tried this and the first draft still came back generic, I’d like to hear what you told it. Responses are open.</em></p><p><em>I’m Maya. I write for TeamBrain about getting a whole small team working with AI, not just one person, and everything here comes from how our own team works, from the project manager to the VAs to the clients. If you want the bigger picture, start with our free guide, </em><a href="https://teambrain.co/free-guide/?utm_source=medium&amp;utm_medium=article&amp;utm_campaign=first-ai-setup-no-tech-team"><em>The New Way of Working with AI</em></a><em>.</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=98a9545f15bd" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[You’re handing AI the big job first. That’s why it disappoints.]]></title>
            <link>https://medium.com/@teambrain/youre-handing-ai-the-big-job-first-that-s-why-it-disappoints-58785dac7f52?source=rss-6854dff92520------2</link>
            <guid isPermaLink="false">https://medium.com/p/58785dac7f52</guid>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[claude]]></category>
            <category><![CDATA[chatgpt]]></category>
            <category><![CDATA[small-business]]></category>
            <category><![CDATA[productivity]]></category>
            <dc:creator><![CDATA[Small Teams Wrangling AI]]></dc:creator>
            <pubDate>Fri, 30 Oct 2026 12:01:02 GMT</pubDate>
            <atom:updated>2026-10-30T12:01:02.137Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*tFePeoMAzB73B48pW8fh4w.png" /></figure><p>Most people hand AI the biggest job on their list first, get something generic back, and decide the tool isn’t ready. The order is usually what went wrong, and you already know the right order from hiring people.</p><p>Picture your next hire on her first Monday. You wouldn’t give her the biggest client account, a blank page and the instruction “grow it.” You’d give her the weekly report, show her last week’s version, and look closely at what comes back on Friday. If it’s good, she gets something bigger the week after. Nobody reads that as a lack of faith in her. It’s how trust gets built when someone is new and doesn’t yet know how you do things.</p><p>Every new chat puts AI in a similar spot: no history with you, nothing written down, no idea how you like things done. And most people hand it the client account. “Write my whole sales strategy” goes in, a tidy, confident, generic document comes out, and it gets filed under “AI isn’t there yet.” Meanwhile the small repetitive work, the stuff a new hire would have been given first, is still being done by hand.</p><h3>Start where a new hire would start</h3><p>A good manager starts someone new on work that is easy to explain, easy to check and safe to get wrong. The big relationship comes later, after the small handoffs have come back right a few times.</p><p>Run AI the same way. Give away the small, safe, repeatable work first and keep the judgment calls close until you trust the handoff. Four questions sort your week into those two piles.</p><h3>Four questions that sort your week</h3><p>Run any task through these. The more yeses, the sooner it should go to AI.</p><p><strong>1. Is it repeatable?</strong> Do you do this often, in about the same shape each time? Weekly summaries, first-draft replies, notes into a clean doc, a call into action items. Repeatable work is worth the one-time cost of explaining it well, because you get that cost back every time it runs. A one-off task that needs fresh judgment is a worse fit early on, since you pay the whole cost of explaining it for a single use. It’s the same reason you’d train a new hire on the report she’ll write every week before the board deck she’ll write once a year.</p><p><strong>2. Is it low-judgment?</strong> Judgment is the part that lives in your head: taste, reading a room, knowing this client hates being called by their full name. Low-judgment tasks have a clear right answer or format instead, like reformatting, summarizing, drafting from a template, or pulling the key points out of a long thread. If you can’t write down how you decide, AI can’t decide for you. It can help you think it through, and the call stays yours.</p><p><strong>3. Can you write down how it’s done?</strong> This is the one people skip, and it decides whether AI feels reliable or hit-and-miss. AI works off what you give it. If “how we do this” only lives in your head, every handoff starts from zero and the output comes back generic, because generic is all it had to go on.</p><p>Our own team learned a version of this onboarding VAs. Every task a VA picks up comes with a clearly defined deliverable, so the person doing it knows what done looks like and whoever checks it knows what to check against. We also record onboarding as video, because written SOPs don’t stick with people: they forget, and some only take things in by watching. AI has the opposite problem. Within a task it uses everything you hand it, and in the next chat it has none of it unless a written version is there for it to read.</p><p>So write it down once: how you like the weekly update laid out, the five phrases your brand never uses, the steps of your intake process. It’s the difference between telling a new hire the same thing forty times and handing her the doc.</p><p><strong>4. Is it low-stakes if it’s wrong?</strong> Ask what happens if the output is bad and nobody catches it. A messy internal draft you’ll read before it goes anywhere is low-stakes. A message sent straight to your biggest client with no eyes on it is a different matter. Early on, hand over work where a mistake costs five minutes of editing. You’d do the same with someone in her first week, and you’d widen the circle as her work kept coming back right.</p><h3>What to hand over first</h3><p>Run the four questions across a normal week and the same kinds of work rise to the top:</p><ul><li><strong>First drafts of anything.</strong> Emails, posts, replies, outlines, proposals. AI drafts, you edit, and you keep the final word. Editing a decent draft is faster than staring at a blank page.</li><li><strong>Summarizing and reformatting.</strong> A long thread into three bullets, messy notes into a clean brief, a call transcript into action items with owners.</li><li><strong>Turning notes into structure.</strong> You brain-dump a project and AI shapes it into steps someone can pick up. This is where a scattered idea becomes work you can hand off.</li><li><strong>Repeatable admin.</strong> The standard client update, the weekly report in the same format every time, the agenda built from last week’s notes.</li><li><strong>Research legwork.</strong> A stack of notes to start from instead of a blank search bar. You still judge what’s true and what matters.</li></ul><p>Each one has a clear input, a target you can describe and a cheap cost of being wrong. When you’re unsure about a new task, check whether it looks like this batch. If it does, it’s first-week work.</p><p>Picture the first one in practice. It’s Monday and the weekly client update is due. You paste in last week’s notes along with the written-down version of how you like the update laid out, and ask for a draft. What comes back needs a few edits: a date to fix, a line that’s too formal for this client. You make them and send it, and you’ve found two things to add to the written version, so next Monday’s draft needs fewer. That’s the same loop as reviewing a new hire’s first reports, only faster.</p><h3>What stays with you for now</h3><p>Some work should stay with you, the same way a new hire doesn’t run the hard conversations in her first month.</p><p>Keep the <strong>relationship calls</strong>: the hard conversation with a client, the feedback a teammate needs to hear from a person, the negotiation where tone carries the whole thing. AI can help you prepare. It shouldn’t be the one in the room.</p><p>Keep the <strong>things you can’t take back</strong>: anything that goes out under your name with no review, anything you can’t undo, anything where a quiet mistake is expensive.</p><p>Keep the <strong>real judgment</strong>: the strategy call, the choice between three directions, the read on whether now is the right time. Use AI to lay out the options and pressure-test your thinking, then make the call.</p><p>“For now” is doing real work in that heading. The line moves as you get better at describing your work and checking the output, the way a new hire’s responsibilities grow once her work keeps coming back right. Let it move when the handoffs have earned it.</p><h3>Where the handoff leaks</h3><p>There’s one place the new-hire comparison runs out, and it’s where most of the value goes missing. After a month, a new hire knows how you like the report, which client is touchy and what you decided last quarter. AI works from what’s in front of it. Close the tab, and unless that context is written down somewhere it can read, the next chat starts from nothing.</p><p>So you get a great result in one chat, close the tab, and start over the next day. Or one person on the team has AI dialed in for a task and everyone else is back at zero. The work got handed off, and the context stayed behind in one person’s chat history.</p><p>Delegating only adds up if the context lives somewhere the next person, and the next AI session, can pick it up: the writing style, how the weekly update gets done, the call you made last month and why. When that’s written down in one shared place, every handoff starts where the last one ended.</p><p>Our founder works this way every day. A day spent on marketing strategy gets saved to the shared workspace, and so do two hours spent on the brand voice doc. Because the AI then gets the same context every time, a VA could drive it directly off that saved work and get the same kind of output. It’s also why the last step of our VA onboarding is connecting the new VA to the same AI tools the rest of us use, so nobody is working outside the shared context.</p><p>This is what TeamBrain does. It’s the shared brain your team and the AI tools you already use both work from, so the context and the process travel with the task instead of living in one person’s head or one lucky chat. You direct the work, and the workspace keeps the thread so nobody has to rebuild it.</p><h3>Give it the weekly report</h3><p>The comparison has a limit worth saying plainly. AI is a tool your team directs, and the trust you build is in your own handoff: how well the task is written down and how quickly you can check what comes back. What carries over from hiring is the order. Small, checkable work goes first, with how-it’s-done written down, and bigger work follows once the small handoffs keep coming back right.</p><p>So start where you’d start a new hire. Pick one boring task: the weekly update, the standard client email, the notes-into-brief job you do every Monday. Write down once how you like it done, hand it to AI, and edit what comes back. Do it again next week off the same notes. That loop will teach you more about using AI than any amount of reading.</p><p>You don’t need to build anything or become technical. You need to get good at handing off work without losing the thread, to AI and to the people around you. Give it the weekly report this Monday, and keep the client account for when the handoffs have earned it.</p><p><em>If you’ve handed work to AI this way and it broke down somewhere, I’d like to hear which question it tripped on. Responses are open.</em></p><p><em>I’m Maya. I write for TeamBrain about getting a whole small team working with AI, not just one person, and everything here comes from how our own team works, from the project manager to the VAs to the clients. If you want the bigger picture, start with our free guide, </em><a href="https://teambrain.co/free-guide/?utm_source=medium&amp;utm_medium=article&amp;utm_campaign=what-to-delegate-to-ai-first"><em>The New Way of Working with AI</em></a><em>.</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=58785dac7f52" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Why your new hire’s AI writes like a stranger to your company]]></title>
            <link>https://medium.com/@teambrain/why-your-new-hires-ai-writes-like-a-stranger-to-your-company-57346da17c44?source=rss-6854dff92520------2</link>
            <guid isPermaLink="false">https://medium.com/p/57346da17c44</guid>
            <category><![CDATA[claude]]></category>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[productivity]]></category>
            <category><![CDATA[chatgpt]]></category>
            <category><![CDATA[workflow]]></category>
            <dc:creator><![CDATA[Small Teams Wrangling AI]]></dc:creator>
            <pubDate>Fri, 23 Oct 2026 12:01:02 GMT</pubDate>
            <atom:updated>2026-10-23T12:01:02.769Z</atom:updated>
            <content:encoded><![CDATA[<p><em>Every AI on your team was onboarded in private, by whoever happened to use it. The fix is one shared workspace, and nobody writes a line of code.</em></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*Oo5EeYovAKUaCkrEJcGzPw.png" /></figure><p>Your team already uses AI, and everyone’s AI knows something different. One person’s ChatGPT has your voice down, another’s Claude knows your pricing, and the new hire’s knows nothing at all, so her first draft reads like a stranger wrote it.</p><p>Think about how each of those AIs learned what it knows. When a person joins your company, somebody onboards them: how you write, what you charge and why, which projects are live and who owns them. Nobody does that for the AI. Each teammate teaches their own one piece at a time, through whatever they happen to paste into a chat, and none of it reaches anyone else.</p><p>If you onboarded people that way, you would notice within a week. Every hire would be trained by whoever sat nearest, from memory, and never told what the rest of the office was told. That is roughly how AI gets set up on a team today. The fix looks a lot like proper onboarding: one shared brain that Claude and ChatGPT both read from, set up once, with no code.</p><h3>How you notice it</h3><p>It shows up in small ways first. Two people answer the same customer question and the answers don’t match. A draft comes back in a voice that isn’t yours. Someone asks the AI about a decision the team made in March and it guesses, because nobody ever gave it the real answer. The new person spends her first week asking things everyone else already knows.</p><p>The AI is doing what the person in the chat told it to do. What’s wrong is that the company’s knowledge is split across separate heads and separate chat histories, and none of it is connected.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*e5OW48P9ZwHOYz4LX-RkFQ.png" /><figcaption><em>Same question, two different answers. Neither person did anything wrong.</em></figcaption></figure><h3>Why a style guide in a folder won’t fix it</h3><p>The obvious move is to write it all down. Put the style guide, the pricing notes and the project list in a shared folder, and ask everyone to paste the relevant parts into their AI.</p><p>When we onboard a VA, we record it as video, either a dedicated walkthrough or the onboarding call itself, because written SOPs don’t stick. People forget what’s in them, and some people only retain what they’ve watched.</p><p>A folder of docs fails the AI in a different way. Inside a chat, the AI works from whatever it’s given. The trouble is it only gets what someone remembers to paste in. The style guide goes in on Monday and gets skipped on Thursday when someone is in a hurry. The pricing note gets pasted in its March version long after the price changed. Every pasted copy drifts from the original, and the folder becomes the SOP nobody reads.</p><p>What fixes it is the AI reading the shared version itself, every time, whoever is asking. That is all a shared brain is: one workspace that Claude and ChatGPT read from directly, so the onboarding happens on every request, whether or not anyone remembers to do it.</p><h3>What goes in the onboarding pack</h3><p>The context a new contractor needs to be useful is the same context an AI needs. Write it once and both get caught up from the same place. It sorts into three kinds.</p><p><strong>Context is the thinking.</strong> Why you priced the way you did, what you tried last year and dropped, who your customer actually is. The reasons behind decisions, so nobody has to guess or ask again.</p><p><strong>Process is how you do things.</strong> Your writing style, the steps for handling a refund, how you like a proposal laid out. The stuff that lives in one person’s head and walks out the door with them.</p><p><strong>Work is what’s in motion.</strong> The live projects, who is on what, and where each thing stands.</p><p>None of it has to be polished. A useful first entry can be as rough as this, typed the way you’d explain it to a new hire over coffee:</p><blockquote><em>Spring newsletter relaunch. Goes to past customers only, not the whole list. We dropped discount codes last time because people started waiting for them. Still undecided on weekly or every two weeks. Sam owns the send, Priya approves subject lines.</em></blockquote><p>That’s a few lines, and it already answers the questions a new person, or their AI, would otherwise go around the office asking.</p><h3>“No code” means copy and paste</h3><p>Most guides for this are written for developers: servers, API keys, a terminal. For a ten-person shop that reads like a wall.</p><p>The plain version is a connector. Its technical name is MCP, and you’ll see that name in Claude’s and ChatGPT’s settings. You don’t need to know what it stands for, any more than you need to know how email gets delivered to send one. You paste one link, click connect, and the AI can read your workspace. In onboarding terms, it’s handing over the login on day one. Nobody writes code and nobody installs anything.</p><h3>The setup, step by step</h3><p>You can get the first project done in a single sitting.</p><p><strong>1. Make a workspace</strong> for the team. This is where the shared brain lives, and one per team is the normal setup.</p><p><strong>2. Brain-dump one real project,</strong> like the newsletter example above. Don’t organize it. The workspace turns the messy version into a project with its context and its tasks. Getting what’s in your head written down is the one part only you can do.</p><p><strong>3. Add your process and writing style.</strong> Style is the big one, because it’s what people most often get wrong when they draft in their own AI. Describe how your team sounds or paste in a few examples that got it right, plus the checklists you repeat.</p><p>Our own team runs on this layer. When our founder spends a day on marketing strategy, or two hours on the brand voice doc, it gets saved to the shared workspace once. Because the AI then gets the same context every time, a VA could drive it directly off that saved work and produce the same kind of output the founder would.</p><p><strong>4. Invite the team</strong> by email. Each person gets their own login, and you set what they can see and change: a watcher reads, an editor adds and changes, an admin runs the project. It’s access by role, the way you’d give a new contractor the folders they need and not the payroll.</p><p><strong>5. Connect Claude.</strong> Copy the connector link from the workspace settings. In Claude, add a connector, paste the link, and sign in to the workspace once to confirm it’s you.</p><p>Then test it with something only your workspace knows. Ask: “What did we decide about discount codes for the spring newsletter, and why?” An AI that isn’t connected gives you a sensible-sounding answer about discount codes in general. A connected one tells you that you dropped them because people started waiting for them. When it answers with your real reasoning instead of a general guess, it’s connected.</p><p><strong>6. Connect ChatGPT</strong> with the same link. As of 2026, adding a custom connector in ChatGPT takes a paid plan (Plus or above, not the free tier), and you have to switch on developer mode in settings first. It’s also still a beta feature. Claude’s side is the more open of the two right now, so if your team uses a mix, expect a little more friction on ChatGPT.</p><p>Once both are connected it stops mattering who uses which tool. Everyone pulls from the same context and the same style, so the work comes out consistent whoever did it.</p><h3>The step onboarding usually skips</h3><p>At the end of our VA onboarding, we connect the VA to the same AI tools the rest of us use, so they aren’t working outside our systems. I think it’s the step most onboarding leaves out, and it decides whether a new person’s AI knows your company on day one or spends the first week guessing.</p><p>Picture your next hire on her first Monday without it. She opens ChatGPT, asks it to draft a reply to a customer asking about pricing, and gets something polite, generic and wrong about what you charge. She fixes it by asking around, which means pulling someone else off their work. That repeats with every draft until she has rebuilt the company in her own chat history.</p><p>Now the same Monday with it. She opens the same tool, already connected to the workspace. The reply comes back with your real pricing, in your voice, with the note about annual plans your team always adds. She inherited the brain instead of rebuilding it.</p><p>The everyday version is quieter. Two people answer a customer the same way without checking with each other. Nobody re-asks a question that was settled in March, because the answer is in the workspace and the AI can read it.</p><p>The onboarding comparison only goes so far. The AI is a tool your team directs, not a colleague. What it shares with a new hire is the material it needs to be useful, and that part you only have to write once.</p><h3>What built-in memory doesn’t cover</h3><p>One caveat, dated on purpose. As of 2026, the built-in memory and Projects features inside Claude and ChatGPT live in one person’s account. They help that person the way a private notebook helps whoever keeps it. They don’t give your whole team one shared place both tools read from. These products change fast, so check where things stand when you read this, but that’s how it works today.</p><h3>Start with one project</h3><p>Don’t try to onboard your AI into the whole business at once. Pick one real project the team is working on this week. Make the workspace, dump what you know about that project into it, add your writing style, and connect one AI. Once a draft comes back in your real voice from a shared source, you’ll know whether it’s worth rolling out to everyone else.</p><p><em>If you’ve tried to get a whole team’s AI working from the same context and it broke down somewhere, I’d like to hear where. Responses are open.</em></p><p><em>I’m Maya. I write for TeamBrain about getting a whole small team working with AI, not just one person, and everything here comes from how our own team works, from the project manager to the VAs to the clients. See how it works at </em><a href="https://teambrain.co/how-it-works/?utm_source=medium&amp;utm_medium=article&amp;utm_campaign=give-your-team-one-ai-brain"><em>TeamBrain</em></a><em>.</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=57346da17c44" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[The real cost of sharing one Claude Project with your team]]></title>
            <link>https://medium.com/@teambrain/the-real-cost-of-sharing-one-claude-project-with-your-team-a7e25fc1961f?source=rss-6854dff92520------2</link>
            <guid isPermaLink="false">https://medium.com/p/a7e25fc1961f</guid>
            <category><![CDATA[claude]]></category>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[workflow]]></category>
            <category><![CDATA[productivity]]></category>
            <category><![CDATA[chatgpt]]></category>
            <dc:creator><![CDATA[Small Teams Wrangling AI]]></dc:creator>
            <pubDate>Fri, 16 Oct 2026 12:01:02 GMT</pubDate>
            <atom:updated>2026-10-16T12:01:02.495Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*POuZph4l_eHodw1dqmkojg.png" /></figure><p><em>It’s a seat for everyone who needs it, and it still leaves out whoever is on ChatGPT. You can get the same shared context without the plan.</em></p><p>You built a Claude Project with your brand voice and your customer FAQ in it, and it finally answers like your business. Then you go to share it with a teammate and find the share button comes with the Team plan.</p><p>Think of the Project as a binder you keep in your own desk drawer. Everything that makes the answers sound like you is in there: the voice notes, the FAQ, the few emails that got it exactly right. Open the drawer and Claude reads the binder before it says anything. Nobody else can open that drawer, and the way to let them in costs more than it first looks. It also solves a slightly different problem from the one you have.</p><h3>What a Project does, and for whom</h3><p>A Project in Claude is a container for one line of work. You give it instructions and reference files (a style guide, a product sheet, past examples), and every chat you start inside it inherits them. Instead of pasting your brand voice into a fresh chat every morning, you open the Project and Claude already has it.</p><p>That is genuinely useful, but only for whoever owns the Project. It’s the binder in your drawer. It works well for you, and to the person at the next desk it might as well not exist.</p><h3>Why Pro and Max can’t share it (as of 2026)</h3><p>On the individual plans, Pro and Max, Projects are built for one person. You can make as many as you like to organize your own chats and documents, but you can’t hand one to a coworker so you both work from it.</p><p>As of 2026, Anthropic puts Project sharing and collaboration on its Team and Enterprise plans. That is their design, and it’s a reasonable one. Team is a paid per-seat plan with a seat minimum (two seats as of late July 2026, though that number has moved before), and everyone who needs the shared Project needs their own seat. Plan details change, so check Anthropic’s current pricing page before you count the cost.</p><p>In binder terms, the way to let someone open your drawer is to move them into the same office and pay for a desk for each of them. For a small shop where a few people already pay for their own Pro subscription, that means paying a second time, per person, for the one thing you wanted.</p><p>None of this is a knock on Anthropic. If your company lives in Claude all day, a Team plan is probably worth it. But there is a second catch. A shared Project only helps the people inside that one Claude account. Anyone on your team who works in ChatGPT is in a different building, and the binder never reaches them.</p><h3>Picture the Monday this is supposed to fix</h3><p>Picture your newest VA on her first Monday. She uses ChatGPT, because that’s what she used before she joined. A customer emails to ask whether annual plans can be refunded, and she asks her AI to draft the reply. It comes back polite, generic and a little wrong, because her AI has never seen your FAQ or heard how you talk. She fixes it by messaging you, which pulls you off whatever you were doing. Then the same thing happens with the next email, and the one after, until she has rebuilt your binder from scratch in her own chat history.</p><p>Now picture the same Monday where her AI could read the binder from the start. She asks the same question and the draft comes back in your voice, with your real refund policy, without her pasting anything in. Nobody had to message you.</p><p>A shared Project can’t give her that second Monday unless she moves into your Claude account and you pay for her seat. That’s the real cost of sharing one Project with your team: a seat for every person, and nothing at all for the ones on another tool.</p><h3>Move the binder onto a shelf everyone can reach</h3><p>Look at what you were trying to share in the first place. What you want is for your team’s AI to answer from the same context you do. The Project was just the container you happened to put it in.</p><p>So take the binder out of the drawer, put it on a shelf everyone can reach, and point each person’s AI at the shelf.</p><p>AI tools reach outside their own chat through a connector. The open standard behind it is called MCP, the Model Context Protocol, and from here on it’s just a connector. A connector lets Claude or ChatGPT read from a source you control: a workspace that holds your brand voice, your FAQ and your how-we-do-it docs, the same material you loaded into the Project.</p><p>The shape looks like this:</p><ul><li>You keep one shared workspace with your voice, process and reference docs in it.</li><li>Each teammate connects their own AI to that workspace once.</li><li>When anyone asks their AI to draft, answer or summarize, it reads the shared context first and works from it.</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*toRINLj6WMOPu1zfPKzusg.png" /></figure><p>Nobody’s Project got shared, but everybody’s AI now reads from the same binder.</p><p>The shelf also beats a shared Project on some practical points. It works whichever AI each person uses, because the binder sits outside both tools. It’s one place to update, so a change to the brand voice shows up in the next answer from anyone’s AI, and you aren’t chasing stale copies pasted into a dozen chats. And it isn’t tied to anyone’s account, so people can come and go, switch tools or stay on free plans while the shelf stays where it is.</p><h3>How our own team uses the shelf</h3><p>This is how our own team works day to day. When our founder spends a day on marketing strategy, or two hours on the brand voice doc, it gets saved to the shared workspace once. Because the AI then gets that same context every time, a VA could drive it directly off the saved work and produce the same kind of output.</p><p>Connecting to the shelf is also the last step of our VA onboarding. We record the onboarding itself as video, because written SOPs don’t stick: people forget what’s in them, and some only retain what they’ve watched. Then, at the very end, we connect the VA to the same AI tools the rest of us use, so nobody is working outside the team’s systems. I think it’s the step most onboarding leaves out, and it decides whether a new person’s AI knows the company on day one or spends the first week guessing.</p><h3>Setting it up</h3><p>No code is involved, and the steps are ordinary.</p><ol><li><strong>Gather what’s in the binder.</strong> Open the Project you built, or think through what you’d want a new hire’s AI to know. Pull out how you sound (the brand voice, the words you use and the ones you don’t), what you do (the FAQ, the product facts, the policies) and how you do it (the steps for the tasks that come up over and over). You wrote most of this already to make the Project work, so this step is mostly moving it.</li><li><strong>Put it on the shelf:</strong> one shared workspace that isn’t locked to one person’s account and can be read through a connector. TeamBrain is one place to do this. It’s a shared workspace your whole team’s AI reads from, and it exposes a connector so Claude and ChatGPT can both reach it. You layer context the way a real team does: global for how the company sounds, then per-project on top.</li><li><strong>Connect each person’s AI once.</strong> In Claude that’s Settings, Connectors, Add custom connector: paste the workspace link and authorize. Claude’s flow is open on every plan, including free. ChatGPT can do the same, but as of 2026 it takes a paid plan, developer mode switched on first, and it’s still a beta feature there.</li><li><strong>Test it with a real task.</strong> Have someone who didn’t build the context ask their AI to “draft a reply to this customer email in our voice.” If the draft comes out sounding like you without them pasting anything in, your VA gets the second Monday.</li></ol><p>The setup is done when a second person’s AI produces your context without being handed it.</p><h3>When the Team plan is still right</h3><p>Be honest with yourself here. If your team runs on Claude all day, everyone deep in their own chats and Projects, the Team plan buys things a shared workspace doesn’t: Claude’s own collaboration, billed and managed in one place, on the supported path. If everyone already works in the same office, paying for the desks makes sense.</p><p>The shelf is for the common case the plan doesn’t fit: a small team, a mix of AI tools, most people on free or Pro, and one binder everyone needs to read. That’s most small teams right now. Start with the binder you already wrote, put it on the shelf once, and have someone else’s AI show you it can read it.</p><p><em>If you’ve tried to share context across a mixed Claude and ChatGPT team and it broke down somewhere, I’d like to hear where. Responses are open.</em></p><p><em>I’m Maya. I write for TeamBrain about getting a whole small team working with AI, not just one person, and everything here comes from how our own team works, from the project manager to the VAs to the clients. See pricing at </em><a href="https://teambrain.co/pricing/?utm_source=medium&amp;utm_medium=article&amp;utm_campaign=share-claude-project-without-teams-plan"><em>TeamBrain</em></a><em>.</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=a7e25fc1961f" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Why your new hire’s AI writes like a stranger to your company]]></title>
            <link>https://medium.com/@teambrain/why-your-new-hires-ai-writes-like-a-stranger-to-your-company-c0104f5ec12b?source=rss-6854dff92520------2</link>
            <guid isPermaLink="false">https://medium.com/p/c0104f5ec12b</guid>
            <category><![CDATA[productivity]]></category>
            <category><![CDATA[chatgpt]]></category>
            <category><![CDATA[ai-workflow]]></category>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[claude]]></category>
            <dc:creator><![CDATA[Small Teams Wrangling AI]]></dc:creator>
            <pubDate>Tue, 06 Oct 2026 05:18:24 GMT</pubDate>
            <atom:updated>2026-10-06T05:18:24.615Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*SLm5DYV269unzRgzpcDbXA.png" /></figure><p><em>Every AI on your team was onboarded in private, by whoever happened to use it. The fix is one shared workspace, and nobody writes a line of code.</em></p><p>Your team already uses AI, and everyone’s AI knows something different. One person’s ChatGPT has your voice down, another’s Claude knows your pricing, and the new hire’s knows nothing at all, so her first draft reads like a stranger wrote it.</p><p>Think about how each of those AIs learned what it knows. When a person joins your company, somebody onboards them: how you write, what you charge and why, which projects are live and who owns them. Nobody does that for the AI. Each teammate teaches their own one piece at a time, through whatever they happen to paste into a chat, and none of it reaches anyone else.</p><p>If you onboarded people that way, you would notice within a week. Every hire would be trained by whoever sat nearest, from memory, and never told what the rest of the office was told. That is roughly how AI gets set up on a team today. The fix looks a lot like proper onboarding: one shared brain that Claude and ChatGPT both read from, set up once, with no code.</p><h3>How you notice it</h3><p>It shows up in small ways first. Two people answer the same customer question and the answers don’t match. A draft comes back in a voice that isn’t yours. Someone asks the AI about a decision the team made in March and it guesses, because nobody ever gave it the real answer. The new person spends her first week asking things everyone else already knows.</p><p>The AI is doing what the person in the chat told it to do. What’s wrong is that the company’s knowledge is split across separate heads and separate chat histories, and none of it is connected.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*DlIIgsRrL-Xmfp84wtH1jA.png" /></figure><h3>Why a style guide in a folder won’t fix it</h3><p>The obvious move is to write it all down. Put the style guide, the pricing notes and the project list in a shared folder, and ask everyone to paste the relevant parts into their AI.</p><p>When we onboard a VA, we record it as video, either a dedicated walkthrough or the onboarding call itself, because written SOPs don’t stick. People forget what’s in them, and some people only retain what they’ve watched.</p><p>A folder of docs fails the AI in a different way. Inside a chat, the AI works from whatever it’s given. The trouble is it only gets what someone remembers to paste in. The style guide goes in on Monday and gets skipped on Thursday when someone is in a hurry. The pricing note gets pasted in its March version long after the price changed. Every pasted copy drifts from the original, and the folder becomes the SOP nobody reads.</p><p>What fixes it is the AI reading the shared version itself, every time, whoever is asking. That is all a shared brain is: one workspace that Claude and ChatGPT read from directly, so the onboarding happens on every request, whether or not anyone remembers to do it.</p><h3>What goes in the onboarding pack</h3><p>The context a new contractor needs to be useful is the same context an AI needs. Write it once and both get caught up from the same place. It sorts into three kinds.</p><p><strong>Context is the thinking.</strong> Why you priced the way you did, what you tried last year and dropped, who your customer actually is. The reasons behind decisions, so nobody has to guess or ask again.</p><p><strong>Process is how you do things.</strong> Your writing style, the steps for handling a refund, how you like a proposal laid out. The stuff that lives in one person’s head and walks out the door with them.</p><p><strong>Work is what’s in motion.</strong> The live projects, who is on what, and where each thing stands.</p><p>None of it has to be polished. A useful first entry can be as rough as this, typed the way you’d explain it to a new hire over coffee:</p><blockquote><em>Spring newsletter relaunch. Goes to past customers only, not the whole list. We dropped discount codes last time because people started waiting for them. Still undecided on weekly or every two weeks. Sam owns the send, Priya approves subject lines.</em></blockquote><p>That’s a few lines, and it already answers the questions a new person, or their AI, would otherwise go around the office asking.</p><h3>“No code” means copy and paste</h3><p>Most guides for this are written for developers: servers, API keys, a terminal. For a ten-person shop that reads like a wall.</p><p>The plain version is a connector. Its technical name is MCP, and you’ll see that name in Claude’s and ChatGPT’s settings. You don’t need to know what it stands for, any more than you need to know how email gets delivered to send one. You paste one link, click connect, and the AI can read your workspace. In onboarding terms, it’s handing over the login on day one. Nobody writes code and nobody installs anything.</p><h3>The setup, step by step</h3><p>You can get the first project done in a single sitting.</p><p><strong>1. Make a workspace</strong> for the team. This is where the shared brain lives, and one per team is the normal setup.</p><p><strong>2. Brain-dump one real project,</strong> like the newsletter example above. Don’t organize it. The workspace turns the messy version into a project with its context and its tasks. Getting what’s in your head written down is the one part only you can do.</p><p><strong>3. Add your process and writing style.</strong> Style is the big one, because it’s what people most often get wrong when they draft in their own AI. Describe how your team sounds or paste in a few examples that got it right, plus the checklists you repeat.</p><p>Our own team runs on this layer. When our founder spends a day on marketing strategy, or two hours on the brand voice doc, it gets saved to the shared workspace once. Because the AI then gets the same context every time, a VA could drive it directly off that saved work and produce the same kind of output the founder would.</p><p><strong>4. Invite the team</strong> by email. Each person gets their own login, and you set what they can see and change: a watcher reads, an editor adds and changes, an admin runs the project. It’s access by role, the way you’d give a new contractor the folders they need and not the payroll.</p><p><strong>5. Connect Claude.</strong> Copy the connector link from the workspace settings. In Claude, add a connector, paste the link, and sign in to the workspace once to confirm it’s you.</p><p>Then test it with something only your workspace knows. Ask: “What did we decide about discount codes for the spring newsletter, and why?” An AI that isn’t connected gives you a sensible-sounding answer about discount codes in general. A connected one tells you that you dropped them because people started waiting for them. When it answers with your real reasoning instead of a general guess, it’s connected.</p><p><strong>6. Connect ChatGPT</strong> with the same link. As of 2026, adding a custom connector in ChatGPT takes a paid plan (Plus or above, not the free tier), and you have to switch on developer mode in settings first. It’s also still a beta feature. Claude’s side is the more open of the two right now, so if your team uses a mix, expect a little more friction on ChatGPT.</p><p>Once both are connected it stops mattering who uses which tool. Everyone pulls from the same context and the same style, so the work comes out consistent whoever did it.</p><h3>The step onboarding usually skips</h3><p>At the end of our VA onboarding, we connect the VA to the same AI tools the rest of us use, so they aren’t working outside our systems. I think it’s the step most onboarding leaves out, and it decides whether a new person’s AI knows your company on day one or spends the first week guessing.</p><p>Picture your next hire on her first Monday without it. She opens ChatGPT, asks it to draft a reply to a customer asking about pricing, and gets something polite, generic and wrong about what you charge. She fixes it by asking around, which means pulling someone else off their work. That repeats with every draft until she has rebuilt the company in her own chat history.</p><p>Now the same Monday with it. She opens the same tool, already connected to the workspace. The reply comes back with your real pricing, in your voice, with the note about annual plans your team always adds. She inherited the brain instead of rebuilding it.</p><p>The everyday version is quieter. Two people answer a customer the same way without checking with each other. Nobody re-asks a question that was settled in March, because the answer is in the workspace and the AI can read it.</p><p>The onboarding comparison only goes so far. The AI is a tool your team directs, not a colleague. What it shares with a new hire is the material it needs to be useful, and that part you only have to write once.</p><h3>What built-in memory doesn’t cover</h3><p>One caveat, dated on purpose. As of 2026, the built-in memory and Projects features inside Claude and ChatGPT live in one person’s account. They help that person the way a private notebook helps whoever keeps it. They don’t give your whole team one shared place both tools read from. These products change fast, so check where things stand when you read this, but that’s how it works today.</p><h3>Start with one project</h3><p>Don’t try to onboard your AI into the whole business at once. Pick one real project the team is working on this week. Make the workspace, dump what you know about that project into it, add your writing style, and connect one AI. Once a draft comes back in your real voice from a shared source, you’ll know whether it’s worth rolling out to everyone else.</p><p><em>If you’ve tried to get a whole team’s AI working from the same context and it broke down somewhere, I’d like to hear where. Responses are open.</em></p><p><em>I’m Maya. I write for TeamBrain about getting a whole small team working with AI, not just one person, and everything here comes from how our own team works, from the project manager to the VAs to the clients. See how it works at </em><a href="https://teambrain.co/how-it-works/?utm_source=medium&amp;utm_medium=article&amp;utm_campaign=give-your-team-one-ai-brain"><em>TeamBrain</em></a><em>.</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=c0104f5ec12b" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[ChatGPT’s memory finally works. Your team still can’t use it.]]></title>
            <link>https://medium.com/@teambrain/chatgpts-memory-finally-works-your-team-still-can-t-use-it-9b2e02de1875?source=rss-6854dff92520------2</link>
            <guid isPermaLink="false">https://medium.com/p/9b2e02de1875</guid>
            <category><![CDATA[productivity]]></category>
            <category><![CDATA[ai-workflow]]></category>
            <category><![CDATA[small-business]]></category>
            <category><![CDATA[chatgpt]]></category>
            <category><![CDATA[ai]]></category>
            <dc:creator><![CDATA[Small Teams Wrangling AI]]></dc:creator>
            <pubDate>Tue, 29 Sep 2026 04:33:41 GMT</pubDate>
            <atom:updated>2026-09-29T04:39:39.443Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*yZNQxLeigAx7EaFTMdpELA.png" /></figure><p><em>It remembers you. Your ops manager logs in and gets none of it.</em></p><p>You explain your business to ChatGPT on Monday: what you sell, who you sell to, how you like things written. On Tuesday you open a new chat and it has no idea who you are.</p><p>Claude does the same thing. So does the version your coworker set up, carefully, on their own account. This is not a bug and you are not prompting wrong. It is how the tools are built, and the reason matters, because it changes what the fix looks like.</p><h3>Every chat starts on a blank page</h3><p>A chat is a blank page. The AI reads what is in that one conversation and answers from it. Open a new one and the page is blank again.</p><p>That is deliberate. If every message dragged along everything you had ever typed, chats would get slow and expensive to run, and old work would bleed into new work where it does not belong. So the tool wipes the page each time.</p><p>The trouble is that your business does not start over every morning. How you write is the same today as yesterday. How you handle a refund does not change because you opened a new tab. The tool starts from nothing, so you keep teaching it the same things.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*VAEz1nKloqCkdV0LQgnIMQ.png" /></figure><h3>“But doesn’t it have memory now?”</h3><p>It does, and this is where most people stop looking. As of 2026, ChatGPT can hold on to things across your chats and Claude can look back at past conversations. Both features are real and both help.</p><p>What they do not do is remember for anyone but you. That is personal memory. It learns your habits from your own chats. It is not a shared record of how your company works.</p><p>Spend an afternoon teaching ChatGPT your writing style and it stays in your account. Your operations manager logs in, gets none of it, and teaches it her version. It is close to yours, but not identical. Now there are two company styles, each locked inside a different person’s private chats.</p><p>Personal memory solves a real problem for one person. It does not touch the structural one, which is that a business is a group of people who need to work from the same information.</p><p>These features also move quickly. Whatever the limits are the week you read this, expect them to have shifted. That alone is an argument against pinning your company’s knowledge to one vendor’s memory setting.</p><h3>A chat log is not a filing system</h3><p>You have a good back-and-forth and the AI finally understands your customer. That understanding now sits in one place: that chat, in your account. It feels like you saved something. You did, but somewhere nobody else can reach and you will not find again in three weeks.</p><p>Team knowledge behaves differently in three ways. Everyone who needs it sees the same copy rather than it being locked to one login. You can find it rather than scrolling back through hundreds of conversations. And there is one answer to “how do we write a subject line” rather than a slightly different one in each person’s history.</p><p>Using chat history as company memory is like using your sent folder as your filing system. It is all in there somewhere.</p><h3>It compounds with every person you add</h3><p>Alone, this is annoying but survivable. You keep a doc somewhere and paste it in at the start of each chat.</p><p>Add people and it multiplies. Five people run five AIs off five pasted copies of the context, and each copy drifts. One person’s ChatGPT thinks you use exclamation points. Another’s is certain you never do. A new hire has nothing, and no way to know whose version is right.</p><p>You do not end up with a smarter team. You end up with five people using AI slightly differently under one roof. Each is a little faster alone, and the team is a little less consistent. That is why “just paste it in each time” breaks the moment there is more than one of you.</p><h3>The fix is a place, not a prompt</h3><p>Once the problem is stated plainly the answer follows. The context is scattered across private chats that get wiped, so put it in one shared place and use AI that reads from that place every time.</p><p>How you talk to customers goes in once. Your product details go in once. How you handle the awkward situations goes in once. Then when anyone on the team asks their AI to do something, it reads that shared context first. Nobody re-explains, nobody drifts, and a new hire’s AI starts already knowing how the company works.</p><p>This is a real category of tool now, and you keep using Claude or ChatGPT either way. You just point them at something shared instead of retyping your business into every conversation.</p><p>The AI is not broken and you are not using it wrong. It has nowhere to put what your company knows. Give it one place to read from and the forgetting stops.</p><p><em>If you’ve tried to solve this and it broke down somewhere, I’d like to hear where. Responses are open.</em></p><p><em>I’m Kai. I run a small company where the project manager, the designers, the VAs and the clients all work with AI, and I write about getting a whole team onto it instead of one person. The free guide to the setup is </em><a href="https://teambrain.co/free-guide/?utm_source=medium&amp;utm_medium=article&amp;utm_campaign=why-chatgpt-forgets-your-company"><em>here</em></a><em>.</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=9b2e02de1875" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Keep Team Knowledge and AI Work Together With TeamBrain]]></title>
            <link>https://medium.com/@teambrain/keep-team-knowledge-and-ai-work-together-with-teambrain-5cccadc8d117?source=rss-6854dff92520------2</link>
            <guid isPermaLink="false">https://medium.com/p/5cccadc8d117</guid>
            <category><![CDATA[productivity]]></category>
            <category><![CDATA[knowledge-base]]></category>
            <category><![CDATA[team-collaboration]]></category>
            <category><![CDATA[project-management]]></category>
            <category><![CDATA[productivity-tools]]></category>
            <dc:creator><![CDATA[Small Teams Wrangling AI]]></dc:creator>
            <pubDate>Sat, 29 Aug 2026 19:25:29 GMT</pubDate>
            <atom:updated>2026-08-29T19:25:29.459Z</atom:updated>
            <content:encoded><![CDATA[<p>AI can help one person finish a task faster. Teamwork becomes harder when everyone uses it separately.</p><p>One person has useful research inside ChatGPT. Another has a draft in Claude. Project decisions are spread across Slack, documents, and calls. When someone new joins the work, they need the same files and explanations before their AI can help.</p><p>TeamBrain gives the team and its connected AI tools one shared place for documents, tasks, decisions, comments, and project context.</p><p>People can continue working with the AI tools they already use. Their AI reads from and writes to TeamBrain, so useful work does not have to remain inside one person’s chat history.</p><h3>AI Work Breaks Down When Context Stays With One Person</h3><p>Most AI tools begin with an individual conversation.</p><p>You upload a document, explain the project, correct the first answer, and add details the AI could not know. By the end of the conversation, the chat contains useful context about the task and the decisions made along the way.</p><p>The next person does not automatically get that context. They may have the final file, but not the reasoning behind it. Their AI starts from another blank chat and asks for the same information again.</p><p>Teams often work around this by copying prompts, forwarding chat links, uploading the same files, or explaining the project in another meeting. That creates several versions of the same knowledge, and none is clearly the current one.</p><p>TeamBrain puts the shared context between the people and AI tools working on the project.</p><h3>What Is TeamBrain</h3><figure><img alt="TeamBrain Thumbnail" src="https://cdn-images-1.medium.com/max/1024/1*gmifLlObBzXwNT_O_1j-XA.png" /></figure><p>TeamBrain is a shared workspace for teams that use AI.</p><p>It stores documents, briefs, notes, tasks, comments, decisions, assets, and other project information. The same workspace can be accessed by team members and by AI clients connected through MCP or TeamBrain’s API.</p><p>TeamBrain is not meant to replace Claude, ChatGPT, or another AI chat tool. People keep using their preferred AI and connect it to the shared workspace.</p><p>A team member can ask their AI to retrieve the latest client information, save a new draft, update a task, or add a note to the correct project. Another person’s connected AI can then work from that updated information.</p><p>The product also includes project management features, so the documents and the work connected to them can stay in the same place.</p><h3>Build the Shared Brain From Existing Work</h3><p>The TeamBrain workflow begins with the information the team already has.</p><p>You can add documents, briefs, notes, deliverables, and other files to a workspace. These files become the source that team members and connected AI tools work from.</p><p>The workspace can be arranged around a project, client, department, campaign, or another part of the business. This allows a company to separate information instead of placing every document and task into one large collection.</p><p>When the team creates new work, it can be saved back to the relevant workspace. A draft produced through one person’s AI then becomes available to the rest of the people who have access.</p><p>The usefulness of the shared brain still depends on what the team puts into it. Old files, incomplete briefs, or conflicting instructions can give people and AI the wrong context. Someone needs to keep the source material current and remove information that should no longer guide the work.</p><h3>Keep Using Claude ChatGPT and Other AI Clients</h3><p>TeamBrain connects to AI clients through the Model Context Protocol, commonly called MCP. It also supports connections through its API.</p><p>The company lists Claude, ChatGPT, and other major AI clients as supported options. Any AI client that supports MCP can connect directly to the shared workspace.</p><p>The setup involves adding a connector inside the AI client, copying the TeamBrain address from the Connect page, and approving access. TeamBrain describes this as a one-time setup rather than a development project.</p><p>After connecting, people continue giving instructions in their usual AI tool. They can ask it to read a project brief, find the latest decision, create a document, or save completed work into TeamBrain.</p><p>Model support can change as AI clients update their connection options. Teams should check the current setup guide for the exact client and account they plan to use.</p><h3>Store Team Knowledge as Markdown</h3><p>TeamBrain stores its core knowledge as plain Markdown.</p><p>Markdown keeps documents in structured text rather than a proprietary page format. AI tools can read and write the content without working through a visual page builder, while people can still follow the headings, lists, and written information.</p><p>This structure also supports data portability. TeamBrain provides one-click Markdown export, allowing a team to download its documents instead of leaving them inside a format that only TeamBrain can open.</p><p>Markdown works well for briefs, notes, decisions, documentation, and written project material. Files and other assets can be attached to the work they belong to.</p><p>The format does not make every document accurate by itself. Team members still need to review what an AI writes before it becomes part of the shared source.</p><h3>Keep the Reasoning Behind Project Decisions</h3><p>A final document does not always explain why the team chose one direction over another.</p><p>A client may have rejected an earlier concept. A deadline may have changed because another task became more important. The team may have removed a feature after discussing cost or technical limits.</p><p>TeamBrain includes a decision log that records changes with the reasoning behind them. Its activity and task history can show what changed, who made the update, and why the decision was made.</p><p>This gives the next person more than the latest file. They can see the context that shaped it and avoid reopening a question the team already settled.</p><p>Connected AI tools can also use that history when working on the next task. A new draft can follow an earlier decision without requiring someone to paste the same explanation into another prompt.</p><p>The decision record is only useful when the reasoning is captured clearly. Important client, financial, legal, or product decisions should still be reviewed by the responsible person before they become the basis for later AI work.</p><h3>Manage Tasks Alongside Documents</h3><p>TeamBrain includes project management features inside the same workspace.</p><p>The Kanban board tracks work from to-do to done. Tasks can also be shown in a list view with the due date, status, and assignee in sortable columns.</p><p>A shared calendar shows deadlines and events. Comments remain attached to documents, while submissions and status changes can appear in the task discussion. Assets and completed deliverables stay with the work they relate to.</p><p>The dashboard gives the team a view of what is moving and what is blocked. TeamBrain’s AI can create or update tasks and summarize project status from the information in the workspace.</p><p>This reduces the need to keep project context in one tool and task status in another. It does not remove the need for ownership. The team still needs to assign the right person, confirm deadlines, and review AI-created task details.</p><h3>Organize Work With Workspaces and Collections</h3><p>Different teams need different boundaries around their work.</p><p>An agency may create one workspace for each client. A startup could separate product, marketing, operations, and fundraising. A consultant may keep each engagement in its own space.</p><p>TeamBrain supports workspaces organized by project, client, department, or another structure. Access can be limited so people only see the spaces they should use.</p><p>Collections provide another way to group related documents, tasks, and assets. A collection could hold everything connected to a campaign, product launch, or client deliverable without mixing it with unrelated work.</p><p>Clear organization matters because connected AI tools rely on the same structure. A well-named workspace and collection make it easier to retrieve the right context and reduce the chance of placing work in the wrong location.</p><h3>Invite Clients and Contractors Without Adding Full Seats</h3><p>TeamBrain separates team seats from guest accounts.</p><p>Seats are intended for people inside the team. Guests are designed for clients, contractors, and other outside collaborators who only need access to selected workspaces.</p><p>The pricing page says guests can read and comment on the work without taking a full team seat. Their access is limited to the workspaces where they have been invited.</p><p>This can help an agency share a client project without exposing its internal work or other client accounts. A company can also invite a contractor to the material needed for one assignment.</p><p>Workspace owners should still review permissions before inviting someone. A connected AI may transmit relevant workspace content to the AI provider when a user requests an action, so access should match what that person is allowed to use.</p><h3>Start the Day With the Current Priorities</h3><p>TeamBrain sends each teammate a morning email showing what is on their plate and what changed.</p><p>The summary is built from tasks and project activity inside the workspace. It can surface current assignments, recent updates, and work that needs attention without requiring the team to collect the information manually each morning.</p><p>The dashboard provides a broader view of active and blocked work. A reviewer can also receive an email when someone submits work for review.</p><p>These notifications can reduce status questions, but they depend on the workspace being current. An unassigned task or missing deadline cannot appear correctly in a daily summary.</p><p>Teams should also decide which updates need a direct conversation. A morning email can show that a decision changed, but sensitive or high-impact changes may still need a call or written approval from the responsible person.</p><h3>Understand How TeamBrain Handles Data</h3><figure><img alt="TeamBrain Demo" src="https://cdn-images-1.medium.com/max/1024/1*RI-RAiOO9biDYiYLNQwX3Q.png" /></figure><p>TeamBrain says Customer Content is private to the team, protected by workspace access controls, and encrypted in transit. Its privacy policy states that Customer Content is not used to train AI or machine-learning models.</p><p>The policy also explains what happens when a connected AI reads or writes to the workspace. The relevant content is transmitted to that AI provider so it can complete the requested action. The provider then handles the content under its own terms and privacy policy.</p><p>Teams using their own API keys are responsible for the connected accounts and usage. Content processed by TeamBrain’s included AI is used to perform the requested action and, according to the company, is not used to train a model on Customer Content.</p><p>TeamBrain stores and processes data in the United States. Companies with client contracts, data residency rules, or compliance requirements should check whether that location and the policies of connected AI providers fit their needs.</p><p>Deleted items stay in trash for 30 days before permanent removal. Users can also export their documents as Markdown and request account deletion.</p><h3>Choose Between Included AI and Your Own Model</h3><p>Every TeamBrain plan includes access to an open model run by the company. TeamBrain describes it as suitable for everyday work such as drafting, summarizing, and filling gaps.</p><p>Teams can also bring their own AI key and connect a supported model. This gives them control over the provider, model quality, usage volume, and charges on that account.</p><p>The amount of included AI differs between Starter, Pro, and Max. The public pricing page does not list the exact usage allowance for each plan, so teams that expect heavy usage should confirm the current limits before choosing one.</p><p>Bringing your own key does not remove the need to check the AI provider’s data rules and billing. The requests run through that provider and use the connected account.</p><h3>TeamBrain Pricing</h3><p>TeamBrain uses plan limits based on seats, workspaces, guests, storage, and included AI rather than charging for every added person without a limit.</p><p>Starter costs $12 per month or $10 per month when billed annually. It includes five seats, five workspaces, 50 guests, and 10 GB of storage.</p><p>Pro costs $24 per month or $20 per month when billed annually. It includes 20 seats, 20 workspaces, 250 guests, and 75 GB of storage.</p><p>Max costs $49 per month or about $40.83 per month when billed annually. It includes 50 seats, 75 workspaces, 1,000 guests, and 200 GB of storage.</p><p>Every plan lists included AI and the option to bring your own key. Larger teams that need SAML single sign-on or an admin audit trail must contact TeamBrain for a custom arrangement.</p><p>Pricing and launch offers can change, so teams should check the current plan page before purchasing.</p><h3>Who Is TeamBrain For</h3><p>TeamBrain is designed for teams that already use AI and need a shared source for the work created around it.</p><p>Founders and small companies can keep company notes, tasks, decisions, and project documents available to the whole team. Agencies can separate client work into different spaces and invite clients or contractors as guests.</p><p>Consultants and freelancers can organize several engagements while keeping briefs, deliverables, and decision history together. Remote teams can use the shared context to continue work across time zones without repeating every update.</p><p>It will be less useful for someone who only needs a private AI chat or a basic file drive. TeamBrain becomes more relevant when several people and their AI tools need to read from and contribute to the same working context.</p><h3>Put Shared Context Between Your Team and Its AI</h3><p>TeamBrain gives people and connected AI tools one place to find documents, tasks, decisions, assets, and the reasoning behind the work.</p><p>The team keeps using Claude, ChatGPT, or another supported AI client. TeamBrain provides the shared layer where completed work and project context can be stored for the next person or AI that needs it.</p><p>You can begin with one project, add the source documents, connect an AI client, and invite the people who need access. Before using it for sensitive work, review the workspace permissions, connected provider policies, data location, and plan limits that apply to your team.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=5cccadc8d117" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>