The architecture around a reliable agent workflow
Tools, state, permissions, retries, and evaluation determine whether an agent can do useful work beyond a successful demo.
Writing
5 articles
Tools, state, permissions, retries, and evaluation determine whether an agent can do useful work beyond a successful demo.
A practical approach to specifications, context, bounded implementation, and verification when building software with coding agents.
Start with a user task, respect the existing system, and design the data flow, validation, and operational behavior around the AI capability.
Reading code, tracing data, debugging, and understanding system boundaries give you the judgment to evaluate what a coding agent produces.
30M+ records, distributed execution, and high-throughput persistence. An engineering approach to systems that survive scale, concurrency, failures, and time.
Want to work through these ideas together?
Explore mentorship