AI-native developer tooling / agent workflows
Brad ZhangFounding engineer and applied AI product builder for devtools, agent systems, workflow reliability, and open-source distribution.
Open to
Remote / Sprints
Archive / Long-form directory
This archive is organized for rereading, not for feed-speed recency. Start with the cover essay, or move backward through recent entries and monthly sections.
Start here
Useful memory is not what survives by default. It is what a system learns to preserve on purpose.
Most agent systems still treat memory as an overflow bucket. When context windows become expensive, people start saving everything elsewhere and call the result memory. But durable memory is not a storage problem first.…
Ongoing · Original Essay · 12 min read
Period
3 entries
Ongoing · Original Essay · 12 min read
A long-form essay on why agent memory should be treated as editorial selection rather than infinite accumulation.
Most agent systems still treat memory as an overflow bucket. When context windows become expensive, people start saving everything elsewhere and call the result memory. But durable memory is not a storage problem first.…
Ongoing · WeChat Import · 10 min read
A systems view of forgetting, compaction, and recoverability in multi-agent workflows.
The third handoff in an agent workflow is often where the illusion breaks. Responsibility becomes vague, context becomes inherited rather than owned, and every participant assumes someone else is still tracking the miss…
Ongoing · X Thread · 9 min read
Serving, latency, and evaluation become clearer when treated as editorial constraints instead of abstract infra metrics.
Inference infrastructure is often described in the language of throughput and cost. Those metrics matter, but they only become meaningful when tied back to visibility: what the user sees, how quickly they see it, and wh…