Claude Code works great out of the box on greenfield projects. On brownfield codebases, the built-in search starts to hurt you. When to use native agentic search, when to add vector or graph search, and how to wire them up as MCP servers.
The original 3 amigos brought product, developer and tester into the room before a story got built. AI first engineering has quietly reinstated the pattern twice over, once for humans, once for agents. Where the hard question sits is in the reviewer role.
Using AI internally is an operational call. Embedding AI in your product is commercial, legal and architectural. What that changes for what you build, what you sell, and how the business runs.
AI adoption is near-universal but value creation is concentrated in a small minority. The gap between pilots and production, the risks nobody is preparing for, and what actually happens when AI-generated code hits your systems.
How to structure engineering teams when AI agents are part of the workforce.
