The EM's guide to AI adoption (without your engineers hating it) (newsletter.manager.dev)

Weave’s EM guide—based on patterns from 400+ companies—lays out a pragmatic, five-level playbook to adopt AI in engineering without alienating your team: start with AI code editors, add background agents, introduce AI-assisted code reviews, measure outcomes, and build a continuous innovation process. The report warns against shotgun tool purchases and token contests; successful teams pick a small set of tools to evaluate (suggested starting pair: Claude Code and Cursor), run a 3–5 person alpha team for 4–5 hrs/week, and use that cohort to drive org-wide rollout and training. The guide’s concrete technical advice is the useful part: create 10–15 specific “rules files” (e.g., “Always use TypeScript interfaces named XProps,” “Never hardcode API keys; use VITE_ env vars”), and deploy Model Context Protocol (MCP) servers (or GitMCP) so tools see your real DBs, APIs and libraries. Use background agents for low-touch commits but give them proper context/permissions, and use AI reviewers to filter obvious issues while humans focus on architecture. Measure quality not vanity metrics: track adoption quality (DAU by team, agent usage), productivity (time-to-first-commit, review cycle time, deployment frequency), quality trade-offs (PR revert rate, bug-time), and ROI ((time saved × hourly rate) ÷ tool cost). Finally, formalize periodic tool evaluation so you keep adopting better models without chasing every shiny object.