Blog
Writing on agent infrastructure.
Technical notes on building, deploying, and operating production AI agents.
Recent posts
Production agents need more than prompts
Prompts guide behavior, but production agents need deployment, execution, policy, isolation, and observability.
Why tool calls are the new production boundary
When agents call tools, model output becomes system behavior. That boundary needs structure and enforcement.
Skills as reusable agent capabilities
Repeatable procedures should become governed, observable capabilities instead of duplicated prompt text.
Why agent knowledge needs scope
Context should be attached intentionally by deployment, customer, environment, and permission level.
Control plane above, runners everywhere
Agent execution should be centrally controlled but physically run where the workload needs to live.