AI that invents, builds and runs ventures
Autonomous is an AI research lab. We research whether AI can run a venture on its own, with humans at the centre.
* A venture is any project that moves humanity forward: a company, a non-profit, or a project for one of the UN Sustainable Development Goals. Profit is fuel, not the goal.
The world has plenty of ideas.
It lacks the power to build them.
Climate, poverty, ageing populations: our biggest problems need far more solutions, tested far faster than people can build them alone. If AI can take an idea all the way to a working venture, we can try many more, learn faster and grow what works.
Too slow today
Most good ideas are never built. A venture takes years, money and people to get going.
Progress needs attempts
We learn most by trying a lot and keeping what works. AI makes every attempt cheap.
Humans at the centre
AI does the work. People decide what matters, and get time back for what makes us human.
Fail. Learn.
Repeat.
This little agent has never seen this game. It doesn't even know it can jump. It tries, falls, and learns from every fall, until it can't be stopped. That is reinforcement learning, the heart of our research.
Live. This is not a video: the dino is really training in your browser right now, starting from zero. Reload the page and it starts over. Think you can do better? Press space.
Try
It acts, mostly wrong at first. Sometimes it tries something new on purpose: that's exploring.
Fall
Every crash is a signal. A failure tells it just as much as a success.
Learn
It updates what it believes. The strip under the ground is its brain: darker means "jump here".
Now make the game
a venture.
Instead of jumping cactuses, our model writes tickets, predicts which ones matter, solves them and learns from the result. Every ticket makes it a little smarter.
The AI writes new tickets.
It picks what has the most impact.
It does the work, or asks a human.
The result becomes experience.
A simulation of how the loop works.
Build 100.
Keep what works.
Like the dino, we learn most from what fails. Building a venture is now so cheap that we can try many, close most, and grow the ones that work.
A simulation: ideas come in, get built, and either go live or get closed.
Quality over
quantity.
Every venture gets the same test flight. Does it get real traction within a month? Does it earn €5K a month within three months? If it doesn't reach the line, we take it offline and learn from it.
The goal isn't thousands of ventures. It's a few that really work. There are no shortcuts: we have been doing this research for over five years, because learning takes time and quality takes time. We don't throw untested models into the world to get rich quickly.
A simulation of our keep-or-kill stages.
Humanity has reached
the boss level.
Climate change, poverty, war, ageing populations. The challenges keep getting bigger, and our generation decides how this game ends: a utopian world, or a dystopian one.
It's not too late. If AI can learn to beat a game, it can learn to take on our hardest problems, and push the human race forward.
Be part of it.
Join the team, or follow along. We keep it personal, short and relevant.
Be the human in the loop.
Venture analysts, developers and ML engineers who want to shape how ventures get built.
One short email a month.
New jobs, open questions from our research and ways to work together. Have an idea? Just reply to it.