Building Is Easy. Growing Is the Hard Part.
Here's How We Grow Kai.

Introduction

This is the first issue of How We Grow, a weekly log of exactly what we are doing to grow Kai. Closing the gap between building it and getting it seen is the job Lambert and I have, and we are going to write down what we try, week after week, while we take Kai from 1,000 people on a waitlist toward something much bigger. We are in closed beta now, with the public launch a few weeks out.

Since it is the first issue, this one is a catch-up. The three months of building that got us to 1,000 signups, what we are doing right now to grow that number, and what is dropping next.

A quick word on the format, since this is the first one. How We Grow is not an interview. It is the two of us, Lambert and me, sitting down once a week to talk through what we actually did: what we shipped, what worked, what flopped, and where we disagree. We run the same job, so there is no expert and no audience. Each week one of us takes the lead and picks the topic, and the next week we swap. Both our names are on every issue. When we see something differently, we say so on the record, because that is the part worth reading.

So here is the other half of this. Lambert runs growth at Kai with me. He is 23, French, finished university last September, and ran a web agency building sites for three years before any of this. We met when he applied for an internship at a previous company. After the first call I wanted to hire him, and after three days I was sure. We have hopped between companies together since. When Kai needed a growth team, he was the first text I sent.

Prefer to watch or listen? Here's the full episode

The first three months: what we built

When the Morgen team decided we were going to build Kai, the marketing brief was wide open and we could build everything from scratch: no CMS, no legacy pages, no brand yet. We spent about three months on the foundation before we turned to growth, and that was the easy part, because almost everyone can build a product now. The calls we made then are the ones everything since has rested on. Three are worth pulling apart: the waitlist, going AI-native on our own site, and the system we built so both keep improving without us.

We almost over-engineered the waitlist

The first thing we built was the waitlist, from scratch. Building it ourselves instead of buying set the tone for everything after, and it is the first thing I would tell another founder to think hard about.

Our first waitlist was too clever.

The very first flow opened with an AI voice call, built on ElevenLabs. You would talk to Kai, and at the end you got a personalized report telling you which tools to use and how to plug them in, plus some free hours. It demoed well. Running it was another story: hard to prompt so it answered real questions, expensive on those voice models, and a wall of friction in front of people who just wanted to leave their email. We sank ten to twenty hours into that flow, then cut it.

We rebuilt the flow five or six times. The version that works now is the simplest one we have ever shipped:

1. You enter your email and save your spot.
2. You get a confirmation email and your number on the list.
3. You get a personal referral link. Every person who joins through it moves you up the line.

That referral mechanic is the one piece of cleverness that earned its place. We are trying to engineer a little bit of virality, and it shows up in the data. One person referred 25 classmates off a single link. When you can see real people pulling other real people onto the list, you keep that feature.

Two more things, both worth stealing:

1. Ask a few light questions right after signup. How big is your team, what is your company, why do you want Kai. That is user discovery you can run before you have a single paying customer, and I would build it in on day zero.
2. Add a "skip the line" path. A small option that lets people book a call with us. Talking to early users is worth more than almost any metric right now, so we made it a first-class option.

The takeaway is simple: do not over-engineer the thing people interact with for ten seconds. If your product is half-baked and you need a waitlist, build the simplest flow that captures an email and creates a reason to share. You can get fancy later, once you know what people do.

Would we build it from scratch again instead of paying for a waitlist SaaS? Yes. The tools exist, but they are expensive, and the day you want to do something they did not anticipate, you cannot.

We went AI-native, and host the site ourselves

On Morgen, our other product, the marketing site lives on Webflow. It works, but we felt the friction. Every change went through the editor, and more and more we were having Claude write custom HTML that we pasted into Webflow anyway. Before that, on my own and on Lambert's agency work, it was WordPress. Always a CMS.

With Kai we made the call right away: build the site ourselves and host it ourselves. It paid off faster than anything else we did, and here is what it bought us.

Speed. We can stand up a new page template or a new content block in about twenty minutes. No CMS structure to fight, no plugin to find.

Performance, by default. WordPress sites pile up plugins. They conflict, updates break things, and the site slows down. We do not have that failure mode. Better still, every time we push to GitHub a Google PageSpeed check runs against the build and blocks the deploy if performance drops. So the site stays fast by default, even on the days we are moving quickly and not thinking about it.

Room to experiment. Because Claude has a knowledge base of our meetings and internal docs, it understands our product and our story better than a fresh contractor would, which makes writing and shipping articles fast. On top of that we can try things a normal blog cannot. We added a "listen to this article" button that runs a text-to-speech model locally and drops the audio straight onto the post. We built an interactive quiz on a comparison article that recommends a tool based on your answers, and it does not always recommend Kai, because that is the honest version. Both keep people on the page longer, which is a real quality signal to Google.

One clarification, because people assume it. We did not hire a website designer. We decided branding was the thing worth investing in, so we brought in a branding specialist who gave us colors and a mascot. That is a story for another issue. Our first version of the site was built before any of that and it was already decent. The brand foundation is what took it up a level, and it gave Claude a clear design system to build against.

Build a system that improves itself

The thread running through all of this is a feedback loop. Every time we make a mistake, we fix it, then we tell Claude not to do it again and add a test or a gate so it cannot happen twice.

The PageSpeed gate is a good example. A couple of weeks ago I pushed a version of the site with some heavier images, and it got slower. So we added the gate that now checks performance on every deploy. The mistake became a rule, and the rule means neither of us makes that mistake again.

The biggest version of this is what we call the blog eval. Before any article goes live, an AI reviews it against 27 rules covering SEO, technical hygiene, design, and performance, and it scores each rule pass or fail. No vague "this looks fine." That is stricter and slower, and it is why we trust it. It is what keeps us publishing quality instead of the random AI-generated articles that never rank. (That eval is worth its own issue, and it will get one.)

If you are building the same way, here is our short list of what to put in from the start:

• A persona. It makes everything downstream, especially content, far more consistent.
• A design system Claude can read. Early on, before we had one, updating a CTA on one page would not propagate, and a week later two pages looked like different websites. A clear system kills that drift.
• A clean folder structure so the AI keeps things consistent instead of making a mess you discover later.
• Automatic image optimization so you can hand it any JPEG or PNG and the format and size get fixed for you.

Every gate like this hands a little more work to the AI and takes it off us. Six months from now we will have gates we cannot imagine today. We do not know what we do not know yet, and that is fine. The point is the loop.

What we are doing right now: lead with the human

Everything up to here is the build: three months of waitlist, website, and the system that keeps them honest. But building it is not the same as growing it, and growing it is the job. So here is the part everyone asks about. What are we doing, right now, to get Kai in front of people?

The goal right now is simple: make noise. Make sure people know the waitlist exists. And the honest version of how is unglamorous. We are on build-in-public every day, posting to TikTok, Instagram, LinkedIn, and Reddit. It takes real time, it does not scale, and it is where almost all of our signups come from. We started Instagram three weeks ago and have already hit 20,000 views. Social is currently about half to sixty percent of our traffic.

In a market where everyone is automating everything and shipping more, the channels that work for us are the ones where we get to be human, so we lean into them. And the format matters more than the effort. We spent a month or two on one heavily-edited video, and it did almost nothing. The clip that performed best was a raw grab from a camera we picked up during a hackathon, just us working and messing around. People connect with the real thing, and the lesson was simple: do not overcook it.

Then build the parts that scale

The problem with human channels is obvious. The day we stop posting, the traffic drops. So in parallel we are building the parts that keep working without us in front of the camera.

SEO, but not yet at full tilt. A brand-new domain starts with almost no authority, so pouring everything into SEO on day zero is a losing fight. We still publish from the start, but the moment to really push is once you start earning organic backlinks, when people talk about and link to your product on their own. We are not fully there, which is exactly why social is carrying the traffic today.

Free tools. Free tools have quietly become one of our main plays. The playbook is always the same: give away far more value than people expect, earn visibility, and convert a slice of it. A week ago we shipped a free meeting transcription tool you can use in one click, no login. Think about that next to Granola, tldv, and Fireflies, companies that have raised tens of millions and charge $50 to $100 a month for meeting transcription. We built a free version in three days during a hackathon.

This week we launch that tool on Product Hunt, aiming for a top-five finish, and next week we ship a Chrome extension that records and summarizes any meeting locally, on your own machine, for free. The bet is the same in both cases: the value spreads itself, and a share of the people who get it come check out what else we are building.

But "anyone can build a free tool now" is the trap, not the win. The hard part is the experience and the path to conversion. Anyone can ship a one-prompt tool. What makes it worth using is the month of iteration after launch, folding real feedback back in. Our tools are alive. We keep improving them.

The three pillars, in one place

If I had to compress everything above into the strategy we are running, it is three things:

1. Lead with the human on social. It does not scale, and it is where the signups are.
2. Make anything you build with AI the best experience people have seen. If it is not, it is noise, and noise does not get seen.
3. Always leave a clear path to conversion. Every social post, free tool, and article should make the next step obvious.

The philosophy underneath all of it: be human. Or really, be a cyborg, an AI-powered human. Build for people, not just for the language models. The slightly imperfect thing is the one people connect with, the same way we like the flawed character in a film. When we over-iterate, we lose that.

What broke

In the spirit of the series, the failures, not just the wins:

• The over-engineered waitlist. The AI onboarding call cost real money and real time, and we cut all of it. Those hours, gone, and the right call.
• The polished video. A month or two of editing, beaten by a raw hackathon grab we barely thought about.
• We are still learning the technical side. Running our own site means deployments and changes we have not done before. Early on we shipped heavier images and dragged performance down, exactly the kind of mistake the PageSpeed gate now catches for us.
• The conversion gap we have not closed. 20,000 Instagram views is nice, but the honest question we keep asking is how much of that converts. Getting even 10 percent of those people onto the site and 2 percent onto the waitlist is the bar, and we are not sure we are hitting it yet. Closing that gap is the next thing to fix.

What's next

That last one is a thread we will pull on in future issues, along with the blog eval, the branding story, and how the free tools convert.

The bigger picture: we are launching at the end of June, and the next few weeks are about lining up the channels that carry it. What is in flight right now:

• A Chrome extension that transcribes your meetings, free. Lambert is shipping a one-click meeting transcription extension that runs right in your browser, no login. Same playbook as the web tool: give away real value, earn visibility, convert a slice of it.
• Team-expansion growth loops and pricing. I am building the PLG loops that let one person pull their whole team into Kai, and working through how we price all of it.
• Three articles a week. We are scaling SEO with a steady cadence: this How We Grow issue, one how-to guide, and one comparative bottom-of-funnel piece (it drives fewer clicks but converts far better), plus the occasional free tool or opportunistic piece.
• Morgen as a bridge. Morgen, our other product, has far more traction and visibility than Kai does today, so we are working out how to use it to put Kai in front of people, both on Morgen's own site and by reaching out to past and current Morgen users.

If you want to follow along, the whole series lives at How We Grow, and the changelog is the receipts of what we release week to week. Join the waitlist and each issue lands in your inbox, along with early access to what we are building.

FAQ