Successful Coding w/AI in Large Enterprises
What actually works when deploying AI coding tools at enterprise scale. Procurement, governance, security, and the adoption patterns that drive real productivity gains.
The Enterprise AI Reality
Enterprise adoption of AI coding tools looks nothing like startup adoption. Where startups say 'just use Claude Code and ship faster,' enterprises face procurement cycles, security reviews, data governance requirements, and the challenge of rolling out tools to thousands of developers with varying skill levels.
Zach Davis and Claire Vo have seen both sides — the excitement of individual developer productivity gains and the reality of organizational adoption. Their conversation cuts through the hype to focus on what enterprises should actually do.
The key insight: enterprise AI coding success isn't a technology problem. It's a change management problem. The tools work. The challenge is getting 5,000 developers to use them effectively, securely, and consistently.
What Actually Works
Start with Champions, Not Mandates
Find 10-20 enthusiastic developers, let them use AI tools freely for a month, then have them present results to their teams. Peer-driven adoption is 3x more effective than top-down mandates in enterprise settings.
Measure Code Quality, Not Just Speed
Enterprise leaders often measure AI success by 'time saved.' Better metrics: code review pass rate, bug density, test coverage, and time-to-merge. Speed without quality creates tech debt that costs more than the time saved.
Create Shared Conventions
AGENTS.md files, prompt libraries, and coding conventions specific to your enterprise. Without shared conventions, every developer reinvents the wheel and AI output is inconsistent across teams.
Invest in Training
A 2-hour workshop on effective AI coding practices produces more organizational value than any tool selection decision. Most developers are underutilizing the tools they already have.
Adoption Patterns
The enterprise adoption pattern that works.
Pilot (Month 1-2)
10-20 developers across 3-4 teams. No procurement formality — use free tiers or individual subscriptions. Goal: prove value and identify use cases.
Expand (Month 3-4)
50-100 developers. Begin procurement process. Create internal guidelines and conventions. Measure productivity metrics against baseline.
Standardize (Month 5-8)
Enterprise-wide rollout. Formal training program. Shared AGENTS.md templates. Integration with existing CI/CD and code review processes.
Optimize (Ongoing)
Continuous improvement of conventions, training, and tool selection. Quarterly reviews of productivity metrics. Feedback loop between developers and leadership.
Security and Governance
Enterprise security concerns about AI coding tools are legitimate but often overblown:
Code privacy: Enterprise plans for Claude, Cursor, and Copilot all offer zero-retention policies. Your code is not used for training. Verify this in the enterprise agreement.
Secret exposure: The real risk — developers accidentally pasting API keys or credentials into AI prompts. Solve with: git-secrets pre-commit hooks, .env file exclusion from AI context, and developer training.
Code review: AI-generated code should go through the same review process as human-written code. Don't create a separate 'AI code review' process — it creates unnecessary friction and implies AI code is inherently less trustworthy.
Compliance: For regulated industries (healthcare, finance), AI coding tools don't change compliance requirements. The output is still code, subject to the same audit and compliance processes as human-written code.
