As far as I know OpenRouter has <100 employees, maybe even <50.

1-2% headcount on data is below average for a tech company but not unbelievable either.

It is true you can get a lot further than before with "just" a clean semantic layer though!

The secret here is that building the semantic layer is a ton of work, predominantly done by data people.

Far less promising companies with zero profits often have thousands more employees…

The Claude-generated README doesn't inspire confidence in that. I believe you checked all the code, but a lot of people are going to see that README and not give the project a second chance.

I run it with a single extension to give it web search access, and then I have a few skills for reviewing and planning. You really don't need that much IMO

I use pi with a ChatGPT subscription. I find that it does better than the native codex harness: - it's very token efficient. I've had whole profiling + refactoring sessions finish in <20k tokens, whereas in codex you can get close to that with just all the cruft OpenAI loads by default before you even send your own messages - the token efficiency means turns are a lot snappier - fine-grained context control by default - I have to disable a bunch of stuff in codex so it doesn't pollute my context with generated memories - I find that OpenAI model sometimes struggle to navigate codex' sandboxing, and burn a bunch of tool calls retrying outside the sandbox when the sandboxed command fails

I gave up on these things thinking that you always needed an API. How does it work with subscriptions?

Codex generally lets you use the subscription with 3rd party agents / harnesses. ON the PI 1.0 you can directly use the openAI log in to hook up your account, so it should be pretty straight forward. I haven't tried myself yet.

Claude subscriptions require usage credits IIRC. OpenAI is more generous and allows the use of the actual subscription.

I disagree. After a bit of experimenting, I actually found Terra to be a very good workhorse model on none-to-medium reasoning, and I actually quite prefer its code to Sol's in many cases. It has less of a complexity to over-complicate things. Where Sol would have a sea of try/except and recoveries for situations that are structurally impossible, Terra would just write nice, sequential code.

Maybe for one-shotting large things Sol is better, but for prod code where I decompose into smaller tasks and read all the code I favored Terra.

Sol 5.6 was still king for architecture/research in my workflow, though.