What the work covers
Claude model family, model routing by cost and task, prompt caching, tool use, agentic workflows, multi-agent orchestration, Model Context Protocol servers, grounding in governed data, evaluation with golden sets, red-teaming, spend controls.
APIs and microservices in Python and FastAPI, background workers and queues, webhooks and real-time streams, fallbacks and circuit breakers.
Agents and integrations
The agents, data and integrations behind that workflow.
Cost and risk controls
Model routing, per-call budgets, guardrails and fallbacks.
Releases
Release plans, testing, monitoring and a rollback for every release.
5AI built into the work
One process at a time, in a sandbox first, with a written list of what the AI reads, suggests and changes. Agents prepare the work; a named person approves each decision before it takes effect. You give the go-ahead for the side-by-side run.
Where it sits: roadmap stage 5