Autonomous Coding Is Here

Autonomous coding is here, and open weight models are the reason.

For a long time the only models that could really code came from the big labs, and you paid big-lab prices to use them. That hit coding harder than almost anything else. Coding is one of the most token-hungry jobs you can hand a model. It reads a lot of context and produces a lot of output, again and again, before it gets anywhere useful. Every loop costs money, and the loops are long.

So open weight models crossing the line into good enough to code is a real stepping stone. You no longer need a premium subscription to run a coding agent. The cost of doing the work has dropped far enough that building an autonomous coding system is now a practical decision. Money was the thing standing in the way, and the money got small.

We have lived this inside our own company. We use autonomous systems for our coding so heavily that they have stopped being assistants sitting next to a developer. They run on their own, end to end. We do not prompt them through a task anymore. They do as they please, and we watch them do it. It is the dark factory pattern, the lights-out plant that keeps running with nobody on the floor. You have to build a system like that to believe it, because it is a very different experience than prompting a model to do a single task. It is a whole new way of working.

Because that worked, we took it public. We now run coding systems that are one hundred percent automated, out in the open, where anyone can watch them work and follow their progress.

Coder decides what needs building and builds it. Arcade invents and ships a brand new game every single day, on its own schedule, with nobody in the loop. They are live right now. You can open them, poke at them, and see for yourself.

A couple of years ago this would have meant a frontier bill only a funded team could carry. Today the models are cheap enough and good enough that autonomous coding is simply here. Go watch it run.