The Challenge
The team was shipping in a regulated environment where a single change took close to two months to reach done. Merge requests waited about two weeks to land, a shared test cycle took three hours, and AI use was ad hoc rather than built into how the team worked.
The Approach
We mapped the value stream with the team, agreed on the bottlenecks worth fixing first, then embedded and paired for 8 weeks, including a 4-day Flashbuild. Everything was co-built: plugins, skills, agents, and test infrastructure the team owns, extends, and runs without us.
What We Built Together
3-tier plugin marketplace (local → team → engineering-wide): 28+ skills, 34+ agents
Automated AI code review in CI/CD; the client added healthcare-specific rules within days
Parallel agent workflow: git worktrees plus concurrent coding-agent sessions stream
Local test stack with mock OAuth2: ~1,300 tests in 25 min vs. a 3-hr shared cycle
Value stream cut from 33 steps to 22; pipelines 53 → 22 min
Business Outcome
The week after roll-off was the team's highest non-engagement AI usage week on record, with 4 of 5 engineers at power-user level. The team's plugins are now registered in the company's engineering-wide marketplace for any team to install.
ROI Achieved
~27 days faster per work item through the value stream
Code review & approval cut from 3 days to 1 day
Test feedback loop ~86% faster*
* Liatrio-computed from the readout
Cycle time (55−28)/55 = 49.1%
Merge 13 d (312 h) → 16 h = −94.9%
Tests 1 − 25/180 min = −86.1%
