Manifestos / design notes / research
Ideas and systems for how agents act, remember, collaborate, and improve.
Personal agents like OpenAI's Dots are the right long-term goal, but now is the wrong time. This essay argues that we should build Agentic Software first, and shows how Sign in with ChatGPT and Claude Mods help.
Intelligence is getting cheaper faster than electricity or compute ever did. Using Xiaomi's open-source MiMo post-training code as an example, this essay shows why the agent era needs a new system for knowledge collaboration.
How Agentic Software carries intelligence into real-world workflows—and how verified experience returns as knowledge, better Agents, and a path toward RSI.
As agents enter every software environment, the missing layer is not another chat interface. It is a governed way for people and agents to collect, review, transfer, and reuse knowledge through real work.
Agentic AI matters not simply because it makes software more intelligent, but because it turns the ability to operate on knowledge—once supplied only by people—into a productive input that can be called, organized, verified, and continuously improved.