The 2025 AI-Startup Pattern
Or: why we see, month by month, another "fastest-growing AI startup" hitting $100+ million in ARR.
Hi, it’s Andreas and I'm back with Growth—my newsletter that breaks down how startups grow through weekly hacks, strategies, and case studies.
Today, I want to dive into why so many AI startups hit past $1 million ARR at lightning speed. I thought, this cannot be a coincidence, even if they are riding a massive hype wave.
And I was right…
So, it feels like every week I read another headline like “fastest-growing AI startup” or “from $0 to $1 million ARR in 12 months.”
This makes it seem like AI startups are killing it and that all you need is AI to build the most successful startup ever.
But you probably know as well as I do that this can’t be true.
All the more reason to take a deep dive and look at what makes Lovable, Mistral, and others so successful.
And surprisingly, there is a pattern.
The AI-Startup Pattern
So, when diving into this, I noticed the following pattern that most of these major startups seem to follow:
Pain-first discovery
Tiny, tool-augmented team
Forward-deployed prototyping
Community launch & private beta
One pain at a time
Let’s have a look.
1. Pain-first discovery
All these exploding tools were built because their founders saw a major pain point or wanted to scratch their own itch.
Most of the AI founders were either second-time founders or had worked in a startup environment - so they knew they couldn’t blindly dive into an idea and start developing it.
Lovable’s founder, Anton Osika, for example, built an early version of Lovable - the open-source GPT-Engineer - while serving as CTO at a startup, to stop his own team from burning hours on boilerplate code. It received 50,000 GitHub stars, so he knew he was onto something.
2. Tiny, tool-augmented team
The era of large startup teams is over. Every AI startup is now built on small, tool-augmented teams.
This means AI startups are leveraging AI tools (lol) to build, scale, and execute faster. This makes the ARR per employee insanely high.
Lovable, for example—even after racing past $20–30M in ARR—the tool employed just 18 people, resulting in over $1M ARR per employee.
3. Forward-deployed prototyping
It seems that these startups never shipped anything that wasn't tested. They embedded their teams directly inside a real customer’s environment (or workflow) to co-build, refine, and validate their AI solutions in real time. And with that, they did the following:
On-site (or deeply integrated) collaboration: The founders or engineers worked closely with users (often literally sitting with them or sharing their daily tools) so they could see pain points firsthand.
Rapid, iterative feedback loops: Instead of spending weeks building a polished demo, they put a minimally viable prototype in front of actual users in days - or even hours. They observed how it was used, gathered feedback immediately, and made tweaks on the fly.
Shared risk and reward: Because they were in the customer’s environment, the customer felt ownership over the solution. They invested time (and sometimes money) earlier, accelerating adoption once the team was ready for a broader rollout.
Real usage data: They didn’t rely on hypothetical “would-you-use-this?” surveys. They collected metrics on actual behavior - number of queries, time saved, integration issues, etc. - which was far more persuasive to investors and future customers.
4. Community launch & private beta
What you see with almost every AI tool is an online community around it - be it on Slack, a forum on the website, or elsewhere.
Within these communities, users are discussing, sharing feedback and learnings, making it a place where real conversation happens.
In some cases, these communities were built before the tool itself and were used to share a private beta. This allowed the founder to iterate based on feedback at lightning speed, accelerating their forward-deployed prototype approach.
5. One pain at a time
All AI tools started out very narrowly, meaning they focused on solving one high-pain problem before expanding their scope and began mostly vertically, then expanded horizontally.
That is, they started by solving one high-pain use case deeply (e.g., “AI agent for contract due diligence” like Harvey AI did). Once they nailed that niche, they exposed underlying capabilities so users could apply them to adjacent workflows (e.g., “risk flagging across all legal documents” or “generating standardized clauses for templates”).
Another notable trend is ecosystem building.
Almost all of the 202x AI tools use third-party integrations - whether through an official app marketplace or community-driven plugins. The more hooks they expose, the more potential revenue streams they unlock (e.g., revenue-sharing on marketplace extensions).
Examples of the AI Startup Pattern
From Harvey AI to ElevenLabs, many of those startups who seems to skyrocket the past year(s) follow this pattern.
Final Words
It seems, from the outside, that it's enough to ride the AI wave and magically generate millions upon millions in ARR with your AI solution.
But that's wrong.
Behind it, there is a strategy that many are executing well—and, of course, a product that truly solves a pain point accelerates all of this.
However, it seems that, compared to the early startup world, today's startup environment is more mature, has learned from the past, and many founders are doing things right.
PS: If you enjoyed this experiment, please tap the like button and let me know in the comments below. Thank you! 💛
Enjoyed this. From my own experience building apps I’ve found getting users to test the app is the most difficult task.
Very good article, Andreas! I often see information about ARR (e.g. revenue), but is there any insight and publicly available information on the overall profitability on such companies?