Is this still useful in 2026?

[platform-kit]

Has this project been abandoned and/or made obsolete by copilot etc?

Does it still perform well compared to, say copilot or claude?

[andreasjansson]

Good question!

In it's current form it's pretty useless. Most of the code was written several years ago, pre-agents (though it has some proto-agentic components that are interesting from an archeological standpoint).

You could probably write a Claude skill to replace all of AutoCog. But since skills don't necessarily trigger when they're supposed to, it would probably have to be a Claude code / OpenCode plugin with tools.

Then there's the issue of running the model. Ideally you'd want to run Claude code on a remote GPU machine so it can actually execute cog predict etc.

I think these complexities highlight the value of a standalone agent that comes pre-packaged with the context, skills, and tools to be an effective automatic cog model builder.

So I plan to revive this and re-release it as open source, but also cloud hosted so you don't have to manage your own GPUs. I think 2026 will be the year of specialized agents :)

Thanks for bringing this up @platform-kit! I will keep this issue open and post again once I've (vibe) coded up AutoCog 2.