---
title: "Why We Lock the Whole Team in a House for a Week"
description: "Issue #008 of How We Grow: 6 days in one house in Caparica with the whole team. What each of us shipped, why we run these 2 months before launch, and the content experiment that turned 400K views into 10 sign-ups."
canonical: "https://hirekai.ai/blog/how-we-grow/008-why-we-lock-the-team-in-a-house"
date_published: "2026-07-20"
last_updated: "2026-07-20"
article_type: "how-we-grow"
author: "lambert"
---
# Why We Lock the Whole Team in a House for a Week

### Key takeaways

- **A hackathon, for us, has no winner.** Nobody competes on the same problem. The whole remote team spends 6 days in one house, everyone owning a different piece, and the only rule is that things ship before the week ends.
- **Run the intense week before the deadline, not after it.** We scheduled this one 2 months out from launch on purpose. Alignment is cheap to fix in July and expensive to fix in September.
- **The best idea came from the leftover time.** Jim finished both of his planned goals in 3 and a half days. What he built with the spare day and a half is the thing the team got most excited about.
- **Views are not a funnel.** 400K views, roughly 70 visits to the site, 10 sign-ups. The content worked as content. It did not work as acquisition, and we are not going to pretend otherwise.
- **Long debates have a ceiling.** Some demos ran 4 hours. We disagree about whether that is a feature or a bug, and this issue keeps both positions instead of resolving them.

## Introduction

This is How We Grow, our weekly log of what we're doing to grow Kai. Last week we were not in our flats: the whole team was in one house in Caparica, just south of Lisbon, for 6 days. This week we're back on our own screens, so this episode is the debrief.

It's our third hackathon this year, and the last one before the September 1 launch. Here's what we cover:

- What a hackathon actually is for us, and why it isn't a competition
- What each of us shipped, including the feature nobody planned
- The content experiment, with the real numbers
- The thing we still disagree about

### Watch or listen

Watch on YouTube: https://www.youtube.com/watch?v=trpGCkupSjg
Listen on Spotify: https://open.spotify.com/episode/0NM6F0kFbUfc107XexzJR3

## What a hackathon is for us

**Nobody is competing.** When Jim tells friends he's off to another hackathon, the first question is always what the prize is. Fair, because that's what the word means: several teams, one problem, one winner. Ours is the opposite. Everyone brings their own problem. The dev team takes backend, frontend, data and AI, David and Danny take UI and UX, and Jim and I take whatever is growth-shaped that week. The one shared rule is that things ship during the week, so there are still a few days left to iterate on real feedback rather than on a demo.

**Why we ran this one now.** We're a remote team spread across countries and in some cases continents, and we get together every couple of months. Jim's read is that the week builds the trust the next 2 months of video calls run on. There was also a specific reason for the timing: we launch on September 1, and being in the same room 2 months out is the cheapest possible moment to find out we're not aligned. Better a hard conversation in a living room in July than over Slack in late August.

**The half of the week that isn't work.** We worked roughly 12 hours a day, and the rest of it mattered just as much. Meals together, the pool, an evening at a restaurant, World Cup games with a Frenchman, someone from the UK and a Spaniard in the same room. I brought my skimboard and the whole team came down to the beach to try it, with predictable results. You learn different things about people there than you do in a standup.

## What the week produced

### The onboarding only 1 in 5 people finished

Jim came in with 2 goals. The first was activation.

Our original onboarding was built quickly, to get the first users through the door, and it did that job. But it had a problem visible from 2 directions at once. In user testing calls, Jim sits with someone while they meet Kai for the first time, and you could feel the friction: long, a lot of tools to connect, and no explanation of why any of those connections mattered. The data said the same thing less politely. Activation was sitting around 15 to 20%, and for a product where nearly everyone signing up is high intent, only 1 in 5 people reaching the end is not normal.

So he rebuilt it with Marco, our CTO. The counterintuitive part: the new onboarding is **longer** than the old one. The change isn't length, it's order. It gives value before it asks for anything, so there's a reason to keep going, and adds a bit of gamification along the way. It's already live. The metrics will tell us in a few weeks whether the instinct was right, and that's a whole episode on its own once we have them.

### The side quest nobody planned

Jim's second goal was collaboration. Kai today is an individual-first product, and the question was what it would look like if a team worked through it: shared tasks, shared meetings, shared briefs. He got through the designs, the concept and an in-app implementation. Worth being precise, because this is the kind of thing that gets overstated: he couldn't touch the backend that week, so those demos ran locally. It's a solid pitch, not a shipped feature. If we prioritize it, it's a 2 to 3 month build.

Both goals were done in 3 and a half days, which left him a day and a half. That leftover time produced the thing the whole team got most excited about.

It's called **do it with Claude**. Kai already holds the context around a task: the meeting it came out of, the email thread behind it, who said what. The feature is a one-click button that hands the task, with all of that context, to another agent to actually execute. Kai gathers, something else does. David and I had the same reaction watching the demo, which was that we hadn't seen another tool do this.

I asked Jim where the idea came from, expecting a story about the room. It wasn't. He'd been circling it for 3 or 4 weeks in a fairly abstract state, and the collaboration work happened to build the foundation it needed. By the time he got to the spare day and a half it was, in his words, just evident.

The reasoning underneath is a positioning bet. More and more work is getting handed to agents like Claude Code and Codex, and more and more teams are building strategic knowledge bases for them to read. What those knowledge bases don't capture is the day-to-day: the email that came in this morning, the thing your colleague said in a meeting, the last Slack message. That's exactly what Kai is good at. So the bet is that Kai becomes the bridge between the strategic layer and the executor. New market, no playbook, which is both the risk and the reason to do it.

**One click, task to agent.**

### Use-case pages, imports, and the first ads

I had 3 things, mostly acquisition with some activation attached.

1. **Use-case pages.** There's a lot you can do with Kai, and we produce a lot of social content about it, but the site never explained the actual flow. So I've been building pages that show a use case end to end. The one I most wanted to explain is the Kai MCP: you connect it to Claude, ChatGPT or Cursor, and then you can be on a beach or in a coffee shop and just ask, out loud, what your next meeting is or what came into your inbox. That's hard to explain in a sentence, which is exactly why it needs a page. 4 are live now, and the fifth ships this week, built around what Jim made.
2. **Importing your data in.** Two flavours. The first is your LLM memory: you can pull what ChatGPT or Claude already knows about you straight into Kai, which matters because Kai's own context comes mostly from email and meetings, a different slice of your life than what you tell an assistant all day. The second is competitor imports, starting with Granola, where the free tier puts your meetings behind a paywall after 30 days. You export them, bring them into Kai, and pick the conversation back up. The end goal is a personalized onboarding for people arriving from Granola that starts the import for them.
3. **Ads.** We now have enough pages and a strong enough design system to build genuinely good ad landing pages, so I spent the rest of the week on those and on getting the campaigns ready. They start this week.

Jim asked whether my week was successful, which is a fair question to turn around on someone. Everything I set out to do got done, so yes. But the unplanned part was more useful: it was the first time I've really worked **inside** the Kai app rather than around it, in the code, working through it with Claude. I now understand how the memory works and how the features connect in a way I simply didn't before. That was never on the goal list, and it's probably what pays off longest.

## 400K views, 10 sign-ups

The other thing we did that week was hire someone to make content.

We'd been thinking about it for about a month. The logic goes back to our very first episode: we want to lead with the human, because it's the one thing AI won't replace. We'd already made videos ourselves with decent enough results to justify going further. But when we film ourselves at a hackathon, we're also the people who are supposed to be building, and the scope collapses under itself.

So we brought in [Lo](https://www.instagram.com/thestroller_lisbon/) for the whole week. She stayed in the house, ate with us, worked alongside us, and filmed both halves of the week. The brief Jim gave her was, more or less, capture the energy. The angle was deliberately top of funnel, brand and awareness, built around funny skits about startup culture and the persona we think is ours (still narrowing that down, but the hunches are the vibe coder, the engineer who lives in Claude Code, and the marketer who does too).

**On volume, it worked.** The target was a minimum of 20 short-form videos. She delivered somewhere between 25 and 30, plus a library of raw footage we'll be cutting from for the next 6 months.

**On reach, it worked.** 2 or 3 of the reels went viral, a couple of hundred thousand views each. Across the week we hit around 400K views and doubled our follower count. They're still performing.

**What "capture the energy" turned into.**

**On the thing that actually matters, it didn't.** Since we started, those views have sent about 70 people to the site, and 10 of them signed up. Jim put it plainly on the call: we hired someone full time for a week, and it should definitely be worth more than 10 sign-ups.

We're not going to dress that up. What we validated is real and worth having: we can make content people genuinely want to watch, we warmed up the algorithm and the account, and we have 6 months of assets. What we did not validate is that any of it moves people toward the product. Those are 2 different skills and we've now demonstrated exactly one of them.

Which gives us a clean next problem: take the hooks and the editing patterns that made these reels work, and aim them at middle and bottom of funnel content. Reach we can now do. Reach that converts is the open question.

## The messier parts of the week

**The demos ran too long, and we don't agree about it.** Every day ends with demo time: everyone shows what they built, and the team argues about it. A couple of those sessions ran 3 to 4 hours.

**Demo time, somewhere in hour 2.**

Jim's position is that there's a thin line between arguing productively and burning each other's afternoon on philosophy, and that a 4 hour session is well over it. Sometimes an hour went on where a button should sit. He doesn't think the debates are the problem, he thinks the absence of a timekeeper is: someone whose job is to say this is enough and make the call. He has a roughly 50 minute attention span and admits that twice he stood up around the 90 minute mark and went to work in the sauna.

My position is softer, for a reason specific to me. I'm not an engineer. I don't have an intuition for how the backend and the frontend hold together, and those long sessions are where I get it. Danny, who got into vibe-coding a few months ago, shows up to demo with 5 or 6 things at once, and following the argument about why one is easy and another is impossible teaches me more about our product than any doc would. Jim's counter, which I think is right: Danny argues by naming real users from calls he's had, the devs argue implementation, we argue distribution, and one answer that satisfies all 3 lenses is genuinely hard to find.

Where we do agree: a lot of that debate was about our own opinions. We'd spend an hour on a sidebar or a CTA with only internal feedback in the room and no user feedback at all. Ship it, watch what happens, then argue with evidence. Our CEO cuts the key decisions when it runs long, but when the whole team is deep in something, it's hard to stop.

**And the caveat worth repeating:** the collaboration work is a concept and a local demo, not something you can use. The onboarding and do it with Claude are live. Those are different states and we'd rather say so than let the excitement blur them.

**The goodies, as promised on the episode.**

## What's next

The summer goal is unglamorous and clear: our first paying users. Alongside that, acquisition, which means being findable on social and on Google. I'm starting to see genuinely green flags on the SEO side, so we'll see where that goes over the next weeks.

The main risk after a week like this is that everything built in the house gets parked and quietly forgotten. So the plan is the opposite: the ads campaigns start this week, the fifth use-case page ships this week, Jim spends the next couple of weeks on go-to-market for do it with Claude, and we watch what the new onboarding does to that 15-20% number. We're also starting to reach out to people who might talk about Kai when we launch.

Oh, and we have stickers now. Jim's summary of that was that we're back to handing out flyers in the street, which is fair, but I'd still like to know how many sign-ups a laptop in a coffee shop is worth.

If you want to follow along, the whole series lives at [How We Grow](/blog/how-we-grow), and the [changelog](/changelog) is the receipts of what we ship week to week. Sign up and each issue lands in your inbox, along with early access to what we're building.

## Sitemap

See the full [sitemap](https://hirekai.ai/sitemap.md) for all pages.