AI native engineering workflows
Give engineers useful context, then make the workflow repeatable.
Decisions that carry across the work.
Problem
AI tools were available, but using them well across planning, implementation, debugging, and review required shared context and practices that teammates could understand and maintain.
My contribution
I developed shared planning context, role-specific agent skills and workflows, persistent memory, and teammate training. I also built an internal debugging MCP server that brought observability and deployment context into coding workflows.
Decision
We kept plans and decisions available across sessions and repositories, with review checkpoints before implementation. We favored a local coding workflow the team could inspect and debug as tools changed. The debugging integration reduced the context gathering needed before investigating failures.
Outcome
The shared workflows and tooling helped the team reduce monthly AI spend by 52%, from $16.2K to $7.8K, while supporting higher delivery volume. Agentic debugging workflows helped the team resolve 123 Jira bugs in three weeks; estimated diagnostic time fell from 30 to 5 minutes. Testing and verification remained a substantial part of delivery.