Adopt Claude + GitHub Copilot · Mid-Market plan

72% of employees use AI without IT approval. Survey which tools, who, and how often.

A 10-dev team saving 1h/day on code review beats 2 devs saving 30min on docs.

Teams often carry 3+ AI subscriptions. Map who pays, what plan, which features are actually used.

They already know what works. Document their prompts, workflows, and workarounds.

Install CLI, add CLAUDE.md, work actual tickets. Track tasks completed, quality, and time saved.

60–70% acceptance is typical. Below 40% means your codebase needs better context (copilot-instructions.md).

Be specific: "No customer PII in prompts. No production credentials. All AI-generated code gets standard PR review."

Verify SOC 2, check where data is processed, confirm training opt-out is default not opt-in.

Enterprise tiers include SAML SSO + SCIM provisioning. Without SSO, you cannot enforce who has access.

AI code has comparable defect rates. The fix is review, not banning AI.

Pre-commit hooks + git-secrets + provider DLP. Prevents API keys leaking into prompts.

Willing teams, mixed frontend/backend. One champion per team tracks adoption.

Tech stack, file structure, naming, testing requirements. Claude Code reads it automatically.

Framework, patterns, naming. Copilot uses it to generate more relevant suggestions.

Build a shared prompt library. Tribal knowledge is the real product, not the tool.

(license cost) / (hours saved × hourly rate). Target payback under one week.

Champions post wins, answer questions, share prompts. Adoption compounds here.

Each wave learns from the last. Wave 1 champions mentor wave 2.

Live-code a real feature with AI. Debugging, tests, refactoring on code they recognise.

Before/after. Time-to-merge, not time-to-first-commit.

Real numbers only. Seats used, API spend, hours saved from developer surveys.

"Would you be upset if we removed this tool?" Over 70% yes means adoption succeeded.