In the past few months, I’ve integrated my team’s processes into AI-driven workflows, moving from manual or error-prone script-heavy tasks to autonomous execution. Here are the core workflows driving the team’s efficiency, plus a few others I’ve added along the way.

Parallelized Development via Automated Worktrees

After scaling my own development with worktrees, I sought to eliminate many of the errors I ran into while manually working with worktrees (e.g., manual configuration errors and branch mismatches). I built a custom Skill that automates worktree setup via a /worktree command, enabling simultaneous multi-branch development while using sparse checkout logic to navigate and isolate our specific service within a 2,200+ service monorepo.

Branch-to-Diff Agentic Pipeline

Our team follows a strict process: branch, change, commit, and diff. Diffs must follow a specific template or the CI build fails. I created a rule that handles this entire lifecycle. It generates the branch, stages changes, and builds the diff using the approved template. It auto-generates high-level titles and context-rich descriptions, reducing the cognitive load on developers and avoiding the overly verbose recommendations usually seen in standard AI models.

Automated JIRA Ticket Management

To reduce the time spent in JIRA, I created a rule that manages the ticket lifecycle via the JIRA REST API. This agent handles the tedious parts of project management:

• Auth & Creation: Authenticates via Uber’s internal system to read and create tickets.
• Transitions: Autonomously moves tickets (Close, Reopen, Move to Review).
• Context: Fetches similar JIRA tickets to help find related historical context.

Synchronized and Updated Documentation

Documentation quickly gets stale; for example, our testing practices haven’t been updated since 2023. I built a rule to synchronize our Confluence “EngWiki” with our codebase. This ensures a single source of truth that stays updated as the code changes. Both developers and agents can now rely on current patterns, helping agents generate the right code on the first try.

Other agent flows

Beyond those core workflows, I’ve added several smaller agent flows for day-to-day tasks:

• On-call alert investigation — Automates initial triage for team alerts: parses alert fields from a URL, searches relevant log namespaces, and produces a structured summary with a diagnosis and recommended next steps. Cuts triage time and improves on-call efficiency across services.