Engineering Manager OS: Master Team Output (–29% PR p50 in 4 weeks)

TL;DR: Outcome: cut pull request (PR) cycle-time median (p50) by 29% in 4 weeks. Action: run the “Manager OS” cadence (daily/weekly/monthly) and measure PR p50 weekly.

Context: New engineering managers often default to “more coding.” I ran a 4-week experiment to test whether small managerial habits—priorities, delegation, feedback, rhythm—could improve whole-team throughput more than an extra manager coding sprint.

What I changed

• Daily discipline (Manager OS)
  • Wrote Top 3 priorities each morning (only tasks I must do).
  • Used delegation prompt: “What do you think we should do?” before giving answers.
  • Delivered 1 micro-feedback per day (positive or constructive).
  • Energy check: blocked 90 minutes for strategy during my peak hours.
• Weekly rhythm
  • 1:1s using 3Q: How are you? What’s blocking you? Where do you want to grow?
  • Team sync (≤30 min): Wins → Priorities → Risks (with named owners).
  • Calendar audit (Fri): killed or delegated one recurring meeting weekly.
  • Weekly reflection (15 min): one tweak committed for next week.
• Monthly system
  • Team health pulse (anonymous, 3 questions).
  • Role clarity: refreshed each person’s top 3 responsibilities.
  • Decision log review: merged, archived, or escalated stale items.
• Definitions used (explicit)
  • PR cycle time: open → first approval or merge, whichever comes first.
  • p50 / p90: median and 90th percentile.
  • WIP (work in progress): open issues/PRs assigned to a person.
  • Planned work %: story points or items tagged “Planned” ÷ total completed.
  • DORA: DevOps Research and Assessment; we reference Deployment Frequency and Lead Time for Changes but tracked PR cycle time as proxy.

Baseline (4 weeks prior, n values shown)

Data window: 4 weeks, n=96 PRs, n=27 deploys.

Observation: Skewed PR times demand medians; outliers were reviewer vacations and unclear ownership.

How to replicate Engineering Manager OS (one-week start)

1. Instrument: Capture 4 weeks of PR cycle times, deploy counts, 1:1 completion, WIP, planned %. Use medians; list outliers with causes.
2. Install daily loop: Top 3, delegation prompt, one micro-feedback, 90-min strategy block at your personal peak.
3. Run 3Q 1:1s this week with every direct; record “blockers → owner → date due”.
4. Team sync ≤30 min: Wins → Priorities → Risks (each risk has an owner and a next date).
5. Calendar audit (Fri): remove or delegate exactly one recurring meeting ≥30 min.
6. Role clarity refresh: write top 3 responsibilities per person; share in team doc.
7. Decision log: list open decisions, assign a DRI (directly responsible individual), due date, and escalation path.

Paste-ready artifact (copy/print)

Engineering Manager OS – Weekly Checklist
Data window to monitor (rolling 4 weeks):
[ ] PR cycle time p50, p90; list outliers + causes
[ ] Deploys/week (median); Lead Time proxy if available
[ ] WIP per engineer (p50)
[ ] Planned work % vs unplanned
[ ] 1:1 completion rate

Daily:
[ ] Write Top 3 (only I can do)
[ ] Ask: “What do you think we should do?” before answering
[ ] Give 1 micro-feedback
[ ] 90-min strategy block during peak energy

Weekly:
[ ] 1:1s (3Q): How are you? Blocking? Growth?
[ ] Team sync ≤30 min: Wins → Priorities → Risks(owner)
[ ] Calendar audit: cut/delegate 1 recurring
[ ] Reflection 15 min: 1 tweak for next week

Monthly:
[ ] Team health pulse (3 Qs)
[ ] Role clarity: top 3 responsibilities/person
[ ] Decision log: close, escalate, or commit dates

Risks & gotchas

• Risk: Engineering Manager drifts back to coding to “help.”
  Mitigation: Protect the 90-min strategy block; track % time on multiplication tasks vs coding.
• Risk: Delegation stalls due to unclear ownership.
  Mitigation: In syncs, every risk gets a named owner and next date; publish in the team doc.
• Risk: PR outliers hide systemic issues (vacations, neglected reviews).
  Mitigation: Track outliers explicitly with causes; add backup reviewers; rotate review ownership.
• Risk: Meetings creep back.
  Mitigation: Enforce weekly calendar audit with a visible count of meetings cut/delegated.
• Risk: 1:1 cancellations reduce signal.
  Mitigation: Set a 90% completion SLO (service level objective); reschedule within the same week.

Results after 4 weeks (n=104 PRs, n=31 deploys)

• PR cycle time p50: 42 h → 30 h (–29%); p90: 120 h → 88 h; outliers reduced from 6 to 2.
• Deployment frequency (median/wk): 5 → 6.
• WIP per engineer (p50): 3 → 2.
• Planned work %: 60% → 72%.
• 1:1 completion: 67% → 93%.

Primary drivers: clearer ownership, faster reviews via named backups, fewer interrupts, and less managerial context thrash.

Next (2025-11-04 target)

• Goal: PR cycle time p50 ≤ 24 h, p90 ≤ 72 h; 1:1 completion ≥ 90%; WIP p50 = 2; Planned work ≥ 75%.
• Actions: introduce “review SLAs” (next-business-day on PRs), expand backup reviewers, and pilot a 2-hour weekly “no-meeting build window” for the team.

Bottom line: A manager improving personal output by 10% loses to a team improving by 30%. The Engineering Manager OS shifts time from “doing” to “multiplying,” and the numbers move accordingly.