Enterprises do not need to build another Codex.
They need to integrate general intelligence into real business processes safely, reliably, and measurably.
Reuse the agent as infra. Engineer the capabilities, controls, and outcomes.
As General-Purpose Agents Get Better, What Is Left to Build?
Many enterprise agent projects still follow a familiar pattern:
Dify or LangGraph
- multiple agent nodes
- tool calling
- RAG
- prompts
- a chat interface
The declining-value skill is connecting multiple agent nodes.
The increasingly valuable skill is knowing:
When should an agent be free to reason, and when must a deterministic system take control?
@lucaronin Very nice try!
But from business perspective, I didn't see why I need to migrate from my obsidian to this new wheel.
If you say open source, there's also other successful modern privacy-first notes, like: https://t.co/KenilEYaNK
With ever stronger of ClaudeCode/Codex/Cursor, now the major debt of software engineering,
is not tech debt anymore,
is understand debt -- the debt to understand AI generated code for you.