The Week Notis Stopped Being a Science Project

For most of the last year, Notis felt like a very promising science project. The product was useful, people loved the idea, and every week I could feel the shape of something real emerging. But useful is not the same thing as durable. Plenty of AI products look magical for five minutes and then disappear into a graveyard of demos, half-baked agents, and screenshots of revenue dashboards that somehow never mention cash flow.

This past stretch felt different. For the first time, Notis reached profitability. We did roughly $15k in revenue over the last 30 days, and we’ve been cash-flow positive since mid-April. I’m not sharing that because I think $15k is some giant victory lap. It isn’t. I’m sharing it because it’s the first moment where the business stopped feeling like a hypothesis and started behaving like a company.

And honestly, that changes how you think. Once you cross that line, even modestly, every product decision gets sharper. You stop optimizing for applause. You start optimizing for trust, retention, and whether the thing survives without needing a mythology around it.

Profitability is downstream of boring things done well

The least sexy explanation for this milestone is also the truest one: we got better at the boring stuff. Reliability improved. Churn started moving in the right direction. LTV improved. Referrals began doing more of the growth work for us. None of this makes for a dramatic founder thread, but that’s the point. Durable software businesses are usually built on a pile of unglamorous fixes that compound quietly until the numbers stop arguing with you.

There’s a version of building in AI where you chase novelty every week. New model, new wrapper, new benchmark, new launch. That game rewards speed, but it also destroys focus. What actually mattered for us was making sure Notis worked more consistently, failed less often, and created enough real value that people kept it in their day-to-day workflow instead of treating it like an entertaining side quest.

That sounds obvious, but the AI market still behaves like everyone is allergic to obvious things. Reliability is still a feature. Predictability is still a feature. A product that quietly completes work every day is worth more than a dazzling one that breaks when you need it.

Why the referral engine started to matter

One of the most encouraging signals lately is that more growth is coming from people telling other people about Notis. I love paid acquisition as much as the next traumatised founder, but referrals are a very different kind of signal. They usually mean the product has crossed from “interesting” to “I’d actually miss this if it disappeared.” That’s the threshold that matters.

When somebody shares software with a friend, they’re putting their own credibility on the line. They’re not recommending a benchmark score. They’re recommending relief. Less mental overhead. Fewer dropped balls. A system that takes things off your plate and actually follows through. If referrals are rising, it usually means the product is creating that kind of relief often enough to become talk-worthy.

The beta that matters more than it looks

We also rolled out model override in beta, which sounds like a niche feature until you spend enough time with serious users. Different models are good at different things. Some are better at fast classification. Some are better at synthesis. Some are better at coding, planning, or handling messy context. Pretending there’s one universal best model is mostly marketing. Giving users more control over where a task goes is a much more honest product move.

I don’t think the future of AI products is hiding every decision behind a fake layer of magic. The right abstraction is not “make the user blind.” The right abstraction is “make the complexity manageable.” Model override is a small step in that direction. It gives people leverage without forcing them to become prompt-engineering goblins.

The bigger shift: from chat-first to UI-first

This is the strategic bet I’m most convinced about now: chat-first AI products are great for demos and often terrible for operational work. A chat box is an elegant universal interface right up until you want state, progress, branching, handoffs, visibility, repeatability, or any of the other boring things that make work feel manageable. Then it starts to feel like trying to run a company from inside a comment thread.

That doesn’t mean chat is useless. It means chat is incomplete. The more agentic a product becomes, the more it needs real interface. Not endless messages. Interface. Boards. Actions. Context. Status. Review points. Memory you can inspect. Systems you can steer without having to re-explain yourself every five minutes.

That’s where I think products like Claude and the broader wave of agent tools still leave a gap. They are powerful, but power presented as a stream of text eventually collapses under its own ambiguity. If AI is supposed to do real work, users need somewhere to see the work, edit the work, route the work, and trust the work. The future is not chat versus UI. The future is AI embedded inside a UI that actually respects how humans run things.

That’s the lens I’m using for Notis now. Messaging-first got us our wedge. It was the fastest path to usefulness. But UI-first is how you build a product people can live inside once the novelty wears off. And if we’re serious about replacing a second job instead of becoming another clever assistant, that transition matters a lot.

A better milestone than revenue alone

Profitability is exciting, obviously. I’m not going to pretend otherwise. But the more meaningful milestone is that the underlying signals are getting healthier at the same time. Better retention. Better economics. Better reliability. More word of mouth. Those are the signals that make revenue worth caring about, because they suggest the business is getting stronger instead of just louder.

There’s still a ridiculous amount left to do. The category is moving fast, expectations are absurd, and every week brings a new wave of products claiming to automate your life while mostly generating more tabs. But for the first time, I feel like Notis has earned the right to be in this conversation as a company, not just an experiment.

And that, for a founder, is a very different kind of energy. Less adrenaline. More conviction. Less theatre. More product. Exactly where I want to be.