Topic · hongwu.xyz
Building with AI Agents
These are my notes from building with AI agents, at work and on my own. The thread that runs through them is context. In “Building Beautiful Context” I tried to define it — the smallest set of high-signal things that make the right outcome plausible — and described the loop we built so a product team and every AI tool it uses work from one curated record. Eight hours at a video agent hackathon taught me the same lesson on real footage: a Context Agent indexed the clips, my intent turned them into a Story, and only then could the Video Agent plan an edit worth watching.
The other thread is the human side of the work. Running up to 16 agents for 30 days, I never ran out of tokens; I ran out of attention, and of the capacity to verify what the agents produced, which is why I want to study what I call Acceptance Engineering for now. Twelve hours inside an AI data center widened the same question: as agents take on more intellectual work, who gets the compute, and how do we keep thinking for ourselves while we use it?
Posts in this topic
- Building Beautiful Context
Where the thread starts: what good context is, and the loop that curates it for a product team and its AI tools.
- Eight Hours to Build a Video Agent. Then I Edited Myself Out.
Context on real footage: a Context Agent, a Story and a Video Agent cut an arcade surprise into vertical shorts.
- I Managed 16 AI Agents for 30 Days. I Became the Bottleneck.
What broke first when I scaled to 16 agents: not tokens, but my attention and my ability to verify the work.
- I Spent 12 Hours Inside an AI Data Center. We’re Worried About the Wrong Divide.
The bigger picture: compute as cognitive infrastructure, and keeping our own judgment while agents do more of the thinking work.