Migrating an existing workflow to Claude Code agents
A safe, incremental playbook for moving an existing workflow onto Claude Code dynamic workflows — strangler pattern, shadow runs, and instant rollback.
Agentic AI, LLM engineering, and the models behind modern automation — multi-agent systems, LLM evaluation and comparisons, RAG, fine-tuning, AI infrastructure, security, and production AI engineering.
From the blog
A safe, incremental playbook for moving an existing workflow onto Claude Code dynamic workflows — strangler pattern, shadow runs, and instant rollback.
Measure dynamic-workflow quality and gate releases with an eval loop — datasets, code and LLM-judge graders, and regression tracking that stops shipping on vibes.
Harden Claude Code dynamic workflows with sandboxing, least privilege, secrets isolation, and layered prompt-injection defense for untrusted input.
Keep Claude Code dynamic workflows cheap and fast with prompt caching, batching, context scoping, and the right multi-agent tradeoff.
Diagnose the three failure modes of Claude Code dynamic workflows — loops, wrong tool calls, hallucinated arguments — with concrete transcript-based fixes.
What to put in context, what to leave out, and why: practical prompt and context design for reliable Claude Code dynamic workflows and agents.
Wire MCP servers into Claude Code dynamic workflows the right way: auth, schemas, error handling, and idempotency for production-grade agent tool use.
Code-level patterns for Claude Code dynamic workflows: structuring goals, small composable tools, layered context, and subagents for reliable agent runs.
Step-by-step: build a dynamic workflow in Claude Code from setup to tools, skills, hooks, and a first end-to-end run an engineer can follow today.