Internal ai-tools Library

Led the buildout of an internal ai-tools library, a tool-agnostic set of reusable agents, skills, rules, and workflows. Gave participating teams a shared starting point for agentic development, codified org-specific patterns, and removed the blank-prompt step from every new agentic project.

The problem

AI adoption across R&D was fragmented. Every team had their own prompts, their own “agent” (usually a single system-prompt file pinned to whatever IDE tool they happened to use), their own conventions. Nothing shared, nothing reusable, nothing visible to leadership. Every team was paying the same onboarding tax in parallel. Any investment a team made was locked to the specific coding assistant they had adopted, so when a better tool landed the work had to be rebuilt.

The approach

• Inventoried existing AI usage across teams and clustered the common patterns
• Codified those patterns as reusable skills, agents, rules, and workflows, deliberately written against the generic Agent SDK + MCP surface so they carry across coding assistants rather than binding to one
• Published the library with clear versioning and a real contribution path
• Partnered with Domain and Squad Architects to drive adoption inside teams

The outcome

• One starting point for every new agentic project in participating teams
• Internal dev workflows (product-compliance scanning for engineering work, automated code review, issue triage) shipped once and reused across teams
• A visible adoption signal for leadership, not just prompt counts

What I’d change

Contribution onboarding deserved more runway up front. Teams that wanted to contribute back hit friction in the first few releases; the program iterated fast, but the path for external contributors was bumpier than it needed to be, and the fix was cheap once it was finally prioritized.