Stop making your prompt do the work your code should be doing
How to architect a chat workflow that stays legible when users answer out of order, switch intent, and ask side questions mid-flow.
TOPIC
Agent architecture
Practical patterns for tools, routing, context, approvals, and agent workflows that remain legible in production.
Articles in this topic
Don't build a second agent
Why your next AI CLI should be a thin client over your real runtime — not a clever local agent that silently drifts from production.
Building a CLI for AI agents. The hard part was the contract.
Why your agent-facing CLI keeps hanging, mis-parsing, and burning time budgets — and the eight-clause subprocess contract that fixes it.
Stop calling your harness a chatbot
Chatbot or agent harness? Learn the architectural difference, choose the right execution model, and avoid building the wrong system.
Stop building flexible AI. Build opinionated AI.
Flexible-everything AI products feel modern and break in production. The frontier vendors already shipped the answer — encode judgment into the default path.
Build an Anthropic agent that survives streaming, tools, and production
A practical four-layer architecture for Anthropic agents that keeps streaming, tools, beta features, and production debugging predictable.