Situation
The team's day-to-day operating tools were behind where they needed to be. The engineering org was mid-migration from Jira to Linear, with workflow plumbing across many repos still pointing at the old system. Code reviews were manual and inconsistent. GTM data lived in three separate operational systems (CRM, product analytics, billing) with no consolidated view. AI tools were available — Claude, agent SDKs — but none of them were integrated into the operational systems where they would compound.
Mandate
Modernize the operating stack. Finish the in-flight Jira → Linear migration. Embed AI where it compounds. Make the team's daily tools match the work they're doing.
Approach
- Map the current operating stack and identify the highest-leverage seams.
- Sequence: finish the in-flight migration first, then layer AI integrations on top of the modernized foundation.
- Architectural decisions favor reuse: every artifact should be a template the team can extend.
What I built and ran
- Jira → Linear migration across the engineering org: managed the cutover end-to-end. Bulk-removed legacy
Refs: <jira>PR-message checks from many repos, set up Jira→Linear autolinks for traceability, worked through verified-signature merge blockers, and unblocked an automation-bot approval issue that had stalled the migration. The modern workflow tool was the foundation everything else got built on. - Multi-agent code review playbook, deployed on a real PR (configuration-server #47). Haiku triage gathers
CLAUDE.mdcontext across the repo; five parallel Sonnet reviewers each take a slice (security, correctness, style, etc.); Haiku scores confidence and posts results as agh pr comment. Code reviews now run with consistent coverage and a tight feedback loop. - GTM dashboard via a custom MCP connector pulling from CRM, product analytics, and billing systems. The team queries it from Claude Desktop and gets live answers instead of waiting on manual SQL pulls. Surfaced findings within the first week: pipeline concentration risk (one deal dwarfing the rest of the pipeline) and a major CRM hygiene issue (only 3 of 392 paying customers correctly labeled in the CRM).
- Agent-ready marketing site: structured data, semantic markup, and the considerations that make the site legible to AI agents acting on behalf of prospects evaluating the company. The site stopped being a thing AI agents had to guess at.
- Reusable AI prompt for marketing-site redesigns: an authored spec the team can feed into Anthropic's Claude Code — a command-line coding tool that drives codebase changes from natural-language instructions. The prompt encodes the architecture choices (Next.js + Ghost CMS, blog integrated, A/B testing) so each redesign runs as AI-assisted iteration on the spec instead of a greenfield engineering project.
Outcomes
- The Jira → Linear cutover completed cleanly across the engineering org; ticket references now resolve in Linear.
- Each AI tool ran in production and was used by the team — not a pilot, not a slide.
- The GTM dashboard surfaced business-critical findings the team hadn't seen — pipeline concentration risk, CRM hygiene gaps, broken data joins between systems.
- Reusable patterns left behind: the multi-agent review playbook is a template for other repos, and the Claude Code prompt makes future site iterations AI-assisted rather than from-scratch.
Why it matters
Most growth-stage teams have one or two pieces of their operating stack stuck in the past — a workflow tool nobody likes, manual processes that should be automated, AI tools sitting unintegrated. Modernizing the stack means understanding where the friction lives and how to make new tools fit, and it usually requires a different engagement for each piece. A Chief of Staff with technical depth moves multiple pieces forward in one engagement instead of queuing them up one consultant or hire at a time.