Two products that are changing the world started as random experiments.

I recently watched 2 videos. A Lenny’s Podcast with Boris, the creator of Claude Code. And a TED talk by Peter, the creator of OpenClaw. I noticed a common theme that goes against conventional startup wisdom.

No problem statement. No business plan. No riskiest hypothesis to validate.

The opposite of what every startup book tells you to do.

1. Claude Code

Boris Cherny, head of Claude Code at Anthropic, on Lenny’s Podcast:

“When I first started Claude Code, it was just supposed to be a little hack.”

He spent a month building weird prototypes. Most never shipped.

His very first test: “What music am I listening to?”

The model figured out how to use a bash tool to answer it.

That was the moment. A toy question.

When he announced it internally, the post got two likes.

https://lnkd.in/gbf-7yUr

2. OpenClaw

Peter Steinberger, a burned-out founder who sold his company and “felt absolutely nothing” for three years, in his TED talk:

“In early 2025, I tried an experiment.”

He built 44 projects in a few months.

One was a WhatsApp bot he took to Marrakesh to navigate.

He sent it a voice message. The agent figured out on its own how to transcribe audio, found an OpenAI key on his laptop, sent the file, got the text back. Nine seconds. He didn’t build any of that.

He put it in a public Discord. It went viral overnight.

Today Jensen Huang calls it the operating system for personal AI.

Neither product started from a problem statement. Neither validated a riskiest hypothesis. Both just played around, and the things they built turned out to be world-changing.

I think this is the part of the AI shift that is under-discussed.

The lean startup playbook assumed building was expensive.

Define the problem. Write the plan. Validate before you build.

Because if you build the wrong thing, you waste months.

But building is no longer expensive.

Boris shipped a bunch of prototypes solo in a month. Peter shipped 44 projects solo in a few months.

When building is near-free, validation becomes the slow path. Exploration is the fast path.

I have been building apps with Claude Code for months now. A working prototype in an hour. Something that used to take a weekend now takes a coffee break.

If you are still writing 10-page business plans before you touch code, you are probably optimizing for a world that no longer exists.

The new playbook is probably: ship weird things, see what people do with them, let the usage tell you what you built.

Boris built a coding hack for himself. Peter built a toy WhatsApp bot for Marrakesh. Neither was expected to gain traction. Both are now.

#AI #Startups #Entrepreneurship #OpenClaw #ClaudeCode