Author here. Every engineering team I have talked to has some version of the same problem: a pile of "we should really fix that" work that never gets prioritized because each item feels too small on its own. A 16-minute CI wait, a snapshot process living in someone's Slack history, a flaky test. None of them ever beat "ship the dashboard redesign" in a planning meeting. But they compound - for our 5-person team, it added up to over 26 hours of dead time per week.
We changed one rule late last year: if something keeps burning team time, it is roadmap work. What made that rule practical was agentic coding - when a CI rewrite takes 20 minutes instead of a day, you can try three approaches and pick the best one. Work that used to sit in the "someday" pile became "this week" work.
Over 90 days we cleared 57 major changesets across 7 workstreams (CI, testing, worktree isolation, ops safety, security, debt, open-source skills) while product delivery continued in parallel. The post covers what worked, what surprised us, and what we got wrong along the way - including an intermediate phase where we made CI faster but the compute bill actually went up.
The thing I would want other teams to take away: the bottleneck right now is not model capability. It is deciding what work is worth attempting. If your team has a hidden queue, the cost of clearing it just dropped by an order of magnitude.
Author here. Every engineering team I have talked to has some version of the same problem: a pile of "we should really fix that" work that never gets prioritized because each item feels too small on its own. A 16-minute CI wait, a snapshot process living in someone's Slack history, a flaky test. None of them ever beat "ship the dashboard redesign" in a planning meeting. But they compound - for our 5-person team, it added up to over 26 hours of dead time per week.
We changed one rule late last year: if something keeps burning team time, it is roadmap work. What made that rule practical was agentic coding - when a CI rewrite takes 20 minutes instead of a day, you can try three approaches and pick the best one. Work that used to sit in the "someday" pile became "this week" work.
Over 90 days we cleared 57 major changesets across 7 workstreams (CI, testing, worktree isolation, ops safety, security, debt, open-source skills) while product delivery continued in parallel. The post covers what worked, what surprised us, and what we got wrong along the way - including an intermediate phase where we made CI faster but the compute bill actually went up.
The thing I would want other teams to take away: the bottleneck right now is not model capability. It is deciding what work is worth attempting. If your team has a hidden queue, the cost of clearing it just dropped by an order of magnitude.
We open-sourced the reusable agent skills that made the gains stick across the team: https://github.com/DiversioTeam/agent-skills-marketplace
Happy to dig into any of the specifics.