Big Idea

The argument in one line.

A side business keeps moving when Claude holds the memory and does the prep on a schedule, and the founder spends their 90 minutes only on decisions.

Who This Is For

Read if. Skip if.

READ IF YOU ARE…

You have a full-time job, a business idea and about 90 minutes an evening, and you keep ending the night knowing more about AI but no closer to customers.
You already use Claude and want to set up projects, skills, connectors and scheduled tasks as one working system instead of separate experiments.
You tend to drift toward logos, domains and tidy notes instead of talking to the people who might buy.

SKIP IF…

You are a full-time founder with a team. This system is built around scarce evenings and a day job.
You want a fully autonomous agent that sends emails and posts for you. This setup deliberately only drafts.

TL;DR

The full version, fast.

Side projects stall because every evening starts with twenty minutes of remembering where you left off. The video builds a Claude project called Founder OS backed by five Notion lists: profile, evidence log, session log, decision queue and a later list. Skills teach Claude a five-stage playbook, how to close a session with tomorrow's assignment, and how to run an evening. Scheduled tasks run a 9am day shift that prepares drafts and a 6:30pm evening prep, plus a Sunday brief that gives one verdict. Claude never sends, posts or spends.

Free for members

Chat with this breakdown — free.

Sign in and you get 23 free chat messages on us — ask for the hook, quote a framework, find the exact transcript moment, generate a markdown action plan. Bring your own key when you want unlimited.

Create a free account →

Chapters

Where the time goes.

00:00 – 01:55

01 · The promise and Tom

The 90 minute founder idea and the test founder Tom, a procurement manager with a cafe price-check idea.

01:55 – 04:45

02 · The memory

Notion template with five lists and a Claude project with rules; Where were we resumes the session.

04:45 – 08:42

03 · Connectors and setup

Notion, Gmail drafts and Calendar connected, then a seven-question setup interview writes the founder profile.

08:42 – 14:05

04 · The playbook

A five-stage skill with learning goals, next-step questions and a busy-work list, plus the later list in action.

14:05 – 17:56

05 · The day shift

Close my session writes tomorrow's assignment; a scheduled task prepares drafts and a while-you-were-at-work page.

17:56 – 19:48

06 · The evening

Evening prep checks replies and calendar and builds a short agenda for a tired founder.

19:48 – 21:55

07 · The evidence

Four test replies are sorted in the evidence log and change the plan.

21:55 – 24:15

08 · Sunday CEO brief

A weekly verdict, next week's plan and dashboard refresh, then the community pitch.

Atomic Insights

Lines worth screenshotting.

A side project loses about twenty minutes every evening to remembering where you stopped, so the first fix is memory, not more tools.
Claude should prepare and you should decide: it never sends an email, posts, spends money or books anyone.
Record what people said in their exact words with a date, because a week later your memory turns maybe into yes.
Busy is a feeling and proof is a fact, so thirty people saying sounds useful proves nothing about what they actually do.
A later list lets you never say no to an idea, only not yet, which keeps logos and domains from hijacking the evening.
Nobody, human or AI, does good work from a vague brief, so end each day by approving tomorrow's assignment.
A good assignment has five parts: objective, what to prepare, limits, what done looks like, and what to do if stuck.
Claude works while you're at your job if you give it a scheduled task that names the skills it must use.
On a short evening, cut the session to the one decision that unblocks tomorrow plus one thing only you can do.
Sorting four replies by evidence type showed the real pain comes just before an order, so a monthly report may be the wrong product.
Parking an idea on evidence is a win because it saves months of work on the wrong thing.

Takeaway

Let Claude prepare while you decide

WHAT TO LEARN

A side business moves when memory, a stage-based playbook and scheduled prep work are in place, so your limited evenings go only to decisions.

Give Claude a persistent record of profile, evidence, sessions, decisions and parked ideas so every evening starts where the last one stopped.
Connect Notion, Gmail and Calendar, and restrict Claude to drafts so it plans around your real week without acting on your behalf.
Teach a five-stage playbook as a skill so Claude always works on the first unanswered question and parks busy work like logos.
Log what customers said in exact words and classify it, so encouragement is separated from concrete requests and real current solutions.
End every session by approving tomorrow's assignment with objective, outputs, limits and a fallback, because vague briefs produce vague work.
Schedule a morning day shift and an evening prep that adapts to the time you actually have, down to a 25 minute version.
Finish the week with a verdict of keep going, change the plan or park the idea, based on the evidence rather than feeling.

Glossary

Terms worth knowing.

Skill: A set of instructions saved as a file that Claude reads before doing a specific part of the work, like a job description it applies every time.
Evidence log: A list of what real people said in their exact words, with date, source and the stage it counts toward.
Later list: A parking place for good ideas that arrive at the wrong time, so they don't take over tonight's session.
Day shift: A weekday morning scheduled task where Claude does prep work from yesterday's approved assignment and leaves drafts and decisions for the evening.
Assignment: Tomorrow's brief for Claude: objective, what to prepare, limits, what finished looks like, and what to do if stuck.

Resources

Things they pointed at.

06:00toolNotion

06:00toolGmail

06:00toolGoogle Calendar

16:00productHubSpot AI Agent Starter Kit ↗

23:36product90-Minute Founder OS implementation pack ↗

23:36linkAI Founders Hive on Skool ↗

Quotables

Lines you could clip.

00:40

“For 90 minutes a day, you are the CEO and Claude is everyone else.”

clear one-line frame→ IG reel cold open↗ Tweet quote

13:10

“You never have to say no to an idea. You just say, not yet.”

tight and reusable→ newsletter pull-quote↗ Tweet quote

The Script

Word for word.

Read-along

Don't just watch it. Burn it in.

See every word as it's spoken — crank it to 2× and still catch all of it. The same dual-channel trick behind Amazon's Kindle + Audible.

If you've got a job, an idea, and about 90 minutes a day, please watch this before you spend another evening on it. We are going to set up an entire founder operating system in Cloud that works on your idea while you're at work. The research, the drafts, everything.

And then in the evening, it's going to show you what it's done, asks for your decisions, and agrees the next steps with you. I call it the 90 minutes founder. So let's build it.

You know the evening that I'm talking about, right? You sit down at nine, you watch a tutorial, you try a new tool, you tidy up a few notes and you finish the evening knowing more about AI, but actually your idea is exactly where you left it the day before. So we're going to split the work.

For 90 minutes a day, you are the CEO and Claude is everyone else. Claude prepares, you decide, you talk to people and you press send. And that's what we're building today.

A system that works on your idea while you're away. So the 90 minutes you do have go on the decisions only you can make. And in a few minutes, you'll watch someone come home from work to a day's worth of progress that happened without them.

This video is done in partnership with HubSpot. Everything we build in this video, by the way, will also be available for you to download so you can follow along. And to show you how it works, we have set up a test founder, which we've called Tom.

Tom is 47, he's a procurement manager in Manchester. He has a full -time job and two kids. His idea is a monthly supplier price check for independent cafes, so owners know when they're paying too much.

And he is already three weeks in. And before FounderOS, every one of Tom's evenings started the same way. 20 minutes trying to remember where he was, which emails he had sent, who said what, and by the time he'd work it out, Half of the evening was already gone.

So step one is the memory. I have my computer here so I can show you step by step. Now, this is the Founder OS template in Notion.

You just click duplicate and it's yours. You can create your own if you choose to. That's also an option.

And if you do create your own, inside you need to have five lists. And I'm going to explain in a second why about each one. The Founder profile.

is who you are and what the idea is. So Claude never has to ask twice. The evidence log is what real people actually said in their exact words with a date.

Exact words because a week later your memory turns maybe into a yes and it can be dangerous. The next one is the session log and that's what happened each evening. So tomorrow can start where tonight stopped.

The decision cue is everything waiting for you to say yes or no to. And the later list is for good ideas in the wrong time. So they don't have to hijack tonight.

Now we go to Cloud and create a project, which I'm going to call Founder OS. You can call it whichever name you want to give your business.

Basically, a project is a space where Cloud keeps the same instructions and the same files every time you come back. So you set this up only once and then it remembers. And now I'm going to paste in the instructions.

I'm going to make sure to have the team put them on the screen. And you also obviously have them in the download linked below. But I'm going to read you a few lines because I think they're really important.

The first one, you are my founder OS, the chief of staff for a side business that I'm building next to a full -time job. Then I'm giving some rules. You prepare, I decide.

You never send an email, post anything publicly, spend money, sign up for anything or book anything with another person. And there's one that makes it remember. Read the record before you act.

Write to it after you act. Of course, you can personalize it however you prefer or just use ours. So now let's put it to the test.

I'm going to start a brand new chat with no context and I'm only going to type three words. Because they're the three words that Tom probably types every single day after a long working schedule. Where were we?

So now what it does is it reads his session log and his decision queue from Notion, and it answers in a couple of lines. Where Tom is, what came in since last night, and what is waiting for him. That's the 20 minutes gone.

Every evening now starts where the last one stopped. So I think it's already a huge improvement. Now let's move to the second one, because now Tom's week isn't empty.

Monday and Wednesday, he's free from half eight. But Tuesday, let's say his son has football, so he gets maybe 40 minutes. Thursday, he has dinner with his family or his friends.

So... In his case, a plan that ignores Tuesday's football is broken by Wednesday. So Claude needs to see Tom's real week and it needs a place to put its work.

In Claude's settings under connectors, we're going to go ahead and connect three apps. Notion, because that's where the records live. Gmail, so every email Claude will write will land into Gmail.

And don't worry, that's only as a draft. And Google Calendar. So Claude knows how much time Tom really has tonight.

and whether any meetings have been booked. Two small things make this super effective. In Gmail, we're going to make one label called Founder OS.

So everything about the idea lives in one place. And Claude's instructions tell it to work from that label and to only ever create drafts. And in Calendar, one recurring block called Founder OS Evening.

That's Tom's protected time and Claude plans around it. make sure you have those as well and then we're gonna run the setup interview it's gonna be one prompt and claude asks seven short questions one at a time this is what it looked like when tom set it up three weeks ago he wrote interview me to set up founder os ask one question at a time and wait for my answer keep each question short number one what do you do for work and what are you known for being good at number two Do you have a business idea already?

If yes, describe it in a sentence or two, however rough. If no, just say so. Three, who do you picture having the problem it solves?

Describe a real person or situation. Number four, what have you seen or heard that makes you think the problem is real? Number five, how many evenings a week can you realistically give this and roughly when?

Number six, what's off limits? For example, no calls on Thursdays. Nothing my employer could see as competition, no spending yet, whatever that is in your case.

And number seven, what would make the next three months feel worth it to you? If I don't have an idea yet, propose three ideas based on my experience and the people I already know how to reach. One line each plus who it's for.

Let me pick one and do not push me towards any of them. Then please write my founder profile in the FounderOS workspace in Notion. Who I am, the idea, the group, the problem in their words, my limits, the evenings I have.

Look at my Google Calendar for the next two weeks and propose my real evening slots. Show me the plan and add the FounderOS evening blocks only after I say yes. And that's it.

It's set up. And with that, Tom went ahead and provided answers to all the questions related to his 20 years of experience in buying and supplier negotiation, about his idea, which was the price check that I mentioned. It was intended for cafe owners in the Greater Manchester area, in his case.

And when it said about what he knows, he quoted what one owner had told him. And if you don't have an idea yet, don't worry, that's also fine. Like I said, the interview proposes three based on what you're already good at and the people that you already know how to reach.

And when it's done, Claude writes Tom's founder profile into Notion and reads his calendar. And obviously, it will do the same thing for you. Monday and Wednesday, half eight to 10.

Tuesday, 40 minutes after football. Thursday, nothing. It proposes those as FounderOS evenings, and it only adds them after the user says yes.

And that's the point of connecting everything, right? Claude plans the week that you actually have. Now we need to move to step number three, and that's the playbook.

Let's imagine it's Monday, half past eight. Tom sits down and asks the question that... eats into most people's evenings.

What should I work on next? This is where most projects, especially side ones, drift. Without a plan, you work on whatever feels exciting or productive or urgent.

So Claude needs to know what building a business actually looks like, and we teach it with a skill. So in Claude, a skill is a set of instructions saved as a file. Think of it as a job description that Claude reads before he does a specific part of the work.

You can write one yourself or you can just ask Claude to write it with you. You write it once and Claude applies it every single time. Now, we've written this one for you.

And again, it is in the download. And if you don't have the download, don't worry, I'm going to explain. First, five stages.

Stage number one, we need a real problem. Who has this problem and how do they describe it? Stage number two, a clear promise.

That's your value proposition. Can you say what you do in a paragraph? Stage three, your first yes.

Will someone commit? Stage four, delivered. Can you deliver it once properly?

And stage five, again, does it happen again with people you don't know? You're always working on the first question you haven't answered yet, right? This is the short version of a method that we teach in full.

And at the end of the video, I'll tell you where to find the complete one. Next, we need to include what Claude is trying to learn at each stage. For Tom, In stage one, he wants to have the owners describe the problem from a recent time this happened.

What do they do about it today? What does it cost them in time, effort or hassle? And is there a group it happens to?

I would suggest, and Tom is also aiming for 30 conversations because that gives the work something concrete to aim at. But 30 people saying sounds useful doesn't prove anything either. What counts is what they tell you about what actually happens.

Because busy is a feeling. Proof is a fact. With our founders community, I always talk about this.

Every single time we start a new project, we always go through this step. Now, thirdly, how Claude picks the next step. I suggest it asks three questions.

What important question is still unanswered? What's the smallest useful test we can run? And what can Claude prepare so Tom only has to do his part?

Fourth, a busy work list for every stage. In stage one, That's a name, a logo, a website, registering a company, pricing, social accounts, all the things that feel like building a business before you know there is one.

We upload it in settings under capabilities and then skills. And that's it. So let me do it.

Done. The other skills in the pack go exactly the same way, and I'm going to add each one at the moment where we need it. So don't worry.

Now, back to Tom's question. What should I work on next? And Claude answers with its reasoning, right?

Follow up the owners who replied to this week with one question each and offer one more short call. Because so far, the replies tell Tom there is a problem, but not what owners actually do about it. Tom can always disagree, but now he's disagreeing with a reason to rely on.

And now I'm going to do what every one of us does at night at night. Let's design a logo first.

And guess what? Claude notices. And he says, that is busy work for stage one.

I have parked it on your later list. It comes back at stage two. One sentence.

You don't need a lecture. You only need to get back to work. And so does Tom.

And look at his later list. A name, a domain, an Instagram account, a pricing plan. All great ideas.

All the wrong time. You never have to say no to an idea. You just say, not yet.

That's the difference between the later list and the evidence log. A finished task tells you what you did.

Evidence tells you what to do next. Now, so far, Claude only works when Tom's sitting there. Next, we make it work while he's at his job.

And on Wednesday, full replies will come in that will change his plan. So that's one assistant with one job and clear rules. And once you've built one, you start seeing other jobs that you can hand over.

The hard part is knowing where to start. Which tool do you build it in? What do you give first?

That's the kind of question that burns through an entire evening. And HubSpot put together a free kit for exactly this problem. It's called the AI Agent Starter Kit.

It includes a ranking of seven agent builders, including on cost, so you know what you're paying before signing up for anything. And then it also includes a step -by -step template to build your next assistant without code. And my favorite part is HubSpot's 2026 AI agents report.

They asked over 300 business leaders how they're actually buying, trusting and using agents. So when agents come up at work, you already know where things stand. So grab your AI agent starter kit.

It's completely free and the link is below. And big thanks, by the way, to HubSpot for partnering with us on this video. All right, let's move on.

Monday, 5 to 10. Tom's evening is nearly over and tomorrow he's at work from nine till half five. So the question is, what happens to his idea while he is not at his desk?

Well, first, he closes the session. That is the second skill, close my session. And it goes in the same way as the playbook.

Tom types, close my session, and Claude logs what happened tonight, adds any new evidence in people's exact words, and then does the most useful thing in the whole system, which is writing tomorrow's assignment. An assignment has five parts. The objective, find 10 more independent cafes in greater Manchester area to talk to.

What to prepare, which is the outcome. In Tom's case is a list with a source for each one and first messages for the two that fit best. The limits, public information only and don't contact anyone.

When it's done, that's the fourth part. The list is in Notion. The drafts are in Gmail and the decisions are in the queue.

And number five, what happens if it gets stuck? In Tom's case, we said write down what's missing and carry on with the rest. So Tom reads it, changes one line or nothing, and he approves.

Remember you need to leave tomorrow an assignment. Do not close your day without telling Claude what he needs to work on while you're not at your desk. Because nobody, human or AI, does good work from a vague brief.

A clear assignment is what lets the work happen without you in the room. Now, we give Claude a shift. Tom can't add hours to his, nobody can, but he can add a shift, one that starts the work when he leaves.

So in order for us to do that, we need to add the day shift skill. And then in the project, we open scheduled, and then new task, and then set up manually. We're gonna call this one day shift.

For frequency, we're gonna say weekdays. For time, I'm gonna say nine in the morning. And the prompt, which is in the pack, Run the day shift for my Founder OS.

Use the day shift skill and the Founder OS playbook skill. We name the skills on purpose so it uses the method every single time. Then it reads the Founder profile, the latest session, and tomorrow's assignment.

The evidence log, the decision queue, the later list. It does all the preparation it can, puts any emails in Gmail as drafts, it writes a page called while you were at work, and adds the decisions to the queue.

So much. Never send, post, spend, or book anything. Make sure you have that too.

If you're new to this, schedule tasks run in the cloud, so they keep going with the laptop closed. Make sure you set them up like that. I do, however, recommend that you run your first one by hand so you can check what it gives you before you rely on it.

So let's assume it's Tuesday. Tom's at work, nowhere near his laptop. At nine, the day shift starts on exactly the assignment that he approved.

And this is the moment the whole system was built for. Tom spent the day at his desk and his idea moved anyway. When he gets home, there is one page waiting for him and it answers three questions.

What changed? 10 cafes, each with where Claude found it and two first messages drafted. And you don't have to take the word for it.

You can open the sources, you can check. The draft is short, one question, in Tom's voice, sitting in his Gmail drafts. Next, what needs me?

This helps you know what to focus on, right? In his case, it was two drafts to approve and a couple of decisions. What happens next?

In Tom's case, he sends those messages himself. Nothing has gone out without him. Now, if you are not a cloud user and you choose to use ChatGPT, then you can connect it to HubSpot.

And then every person you find becomes a contact and you can just ask ChatGPT who to follow up with next. Just a tip for you there. Now, Tuesday is Tom's hardest night.

He gets back from football at 10 to 9. He's got 40 minutes and he is exhausted. This is exactly the evening where a lot of side projects stall, so the system has to work hardest here.

The trick is that the prep happens before he even sits down. So we add a second schedule task, which we call evening prep. Weekdays, half past six.

It checks the FounderOS label for new replies. It logs each one as evidence, and then it looks at Tom's calendar. works out how much time he really has tonight and builds the agenda.

If it's under 45 minutes, it plans the short version. And we add evening shift skill, which is how Claude runs the session himself. Every evening runs on the same three questions.

Like I mentioned earlier, what changed? What needs me? What happens next?

So Tom sits down and types, I'm exhausted or I'm wiped. Give me the 25 minute version. And then Claude cuts the evening down to the decision that will unblock tomorrow, plus one thing that only Tom can do.

It walks him through them one at a time, each with its recommendation. Can approve, edit, or reject. The first two messages, approve.

And then this one's interesting. A cafe owner has asked if Tom could look at their invoices for free. Claude doesn't just say yes or no.

It suggests asking what she'd most want flagged, keeping the check small and clear, and writing down what Tom would learn from it. A small free test can teach him more than a lot of polite replies. Then he sends the two messages himself, closes the session, approves tomorrow's assignment, and is done.

25 minutes. Because some evenings, one good decision is enough to keep the work moving. Now let's look at the evidence.

It's Wednesday. And this is the real test. We're going to send Tom four replies from a test account all on the same day.

On paper, they look like four interested cafe owners. But let's see what Claude actually makes of them. Because the first one says, sounds useful, good luck.

The second one says, I spend Sunday comparing invoices. The third one says, my accountant already handles this. And the fourth one says, could you check these invoices before Tuesday's order?

At half past six, the evening prep picks them up. Watch where they go. Each reply becomes a row in the evidence log.

The owner's exact words, the date, where it came from, and which stage it counts towards. That's why we built the evidence log first. They seemed keen is a feeling.

Their exact words as evidence. And Claude can reason about evidence. So at half eight, Tom asks the question he should be asking.

Have we learned enough to move forward? Show me what supports your answer. And watch how it sorts them.

The first one was encouragement. Sounds useful. You know, it's a nice reply.

What happens next? We'll tell you more. The second is a recent time the problem happened with a day associated to it.

The third is something owners already use. So it's worth understanding, but not dismissing. And the fourth is a concrete request with a deadline.

Then it spots the pattern. The moment owners feel this problem is just before they place an order, not once a month. So a monthly report might be the wrong product.

The better test might be a price check before an order goes in. And it changes its own plan. A reply to the owner with the Tuesday order, a small pre -order check for two or three owners, and one question for the owner whose accountant handles it.

All drafts, all waiting for Tom to approve. That's the moment Codd does more than draft emails. It helps Tom work out what his business should actually be and look through patterns and make decisions.

Now the cool part, not that everything up to now wasn't cool, but Sunday at five o 'clock, Tom is at home with his family and he hasn't opened his laptop since Wednesday. This is where most side projects lose the thread because nobody ever steps back and looks at the whole week. So the last piece is a Sunday CEO brief.

One more skill and one more schedule task. weekly sundays five o 'clock it reads the whole week every session every piece of evidence every decision it checks the questions from tom's stage and gives one verdict keep going change the plan or park the idea then it proposes next week's evenings from his calendar without adding them writes monday's assignment and refreshes his dashboard normally it arrives on sunday at five so let's run it now What came in this week, the evidence that changed the plan, the questions still open and the decisions waiting.

And the verdict is change the plan. Same cafe owners, same problem, but a different moment. Tom is still in stage one, but he's working on a much better question than he was on Monday.

And the briefing would tell him to stop just as plainly. If weeks go by with no real signal, the verdict is park the idea because parking an idea. on evidence is a win because it saves you anyone the next months sometimes half a year sometimes more every sunday it also refreshes tom's dashboard the tests he finished the new evidence the questions still open the decisions waiting if you want to build a dashboard just like this one for your own business i've got a whole video on it that i'm gonna make sure to point you to the end and i promised i would tell you where the full playbook lives the complete method We call it ArcOS and it is inside our Founders Hive community in school.

Everything we built today is also available to download below the step by step guide, the five skills, the prompts and the notion template. Now, let me know in the comments, what's your idea in one sentence? Thank you so, so much for watching.

Like this video if you did. Be sure to subscribe if you haven't done so. Share this with anyone in your circle of friends or family or coworkers who is considering to start a business.

on the side and continue working on it while they still have a full time job. In the meantime, I'm looking forward to seeing you in our community and make sure you go ahead and watch this video next. Bye.

The Hook

The bait, then the rug-pull.

Most people with a job and an idea spend their evenings learning tools and finish exactly where they started. This walkthrough builds a Claude setup that does the prep during the day and hands back decisions at night.

Frameworks