OpenAI’s new GPT-5 Codex is impressive at coding tasks. But, in a huge codebase, intelligence isn’t the main bottleneck—context is.

Been saying this. We don't need AGI, we need ACI. Artificial Contextual Intelligence. To be AGI it requires ACI, imo.

I work with customers who have codebases that are millions and tens of millions of lines. They need an OSS AI memory and context layer like @MemMachine_ai

100%! Once you move from POC and simple code, you’re forced to break your code base up into modules that fit the context window. You can also use AI to generate a framework for assembling the modularity. It works!

Context management + semantic/symbolic retrieval is key for high quality agents

does it modulate itself ?

we've been saying this for years but everyone kept throwing bigger models at the problem. turns out rag and context windows matter more than another trillion parameters who knew

I created a `fid` command that uses a little-known ability of `find` to create a digest of source files, ignoring binary files to speed up grep. It searches the whole codebase as if it was a single mmap'd document.

it's not impressive at coding tasks. engineers spend 20% more time on code generated openai. give this shit a rest

So true

ELON MUSK, THIS IS WANDA PITTS GARY. I HAVE YOUR # 618. AND I PURCHASED A TESLA MEET AND GREET. NEED ADVICE DO YOU DEAL WITH WHATSAPP. I HAVE MORE INFO

Interesting. But how are you handling codebase/data security?

I've built a system context file into my code base - I generally ask the AI to run a new thread and run as long as I can before it maxes the context window and update the context. What's the benefit of using your app ? Also 600 msgs seems low?

Code is always about context.

maybe the Evolutive aspect of the Context is Needed to perform Solving