1️⃣ Test the office rule against retention A 2024 Nature paper on a Trip.com trial found that two home-working days a week reduced resignations, with no detected damage to performance. It tested hybrid work, not fully remote work.
💡 Why it matters Pilot the schedule against output and retention. Protect mentoring time rather than treating office attendance as performance.
☕ Coffee talk Which retention problem is your five-day office rule supposed to solve?
2️⃣ Measure AI time savings before budgeting them In METR’s early-2025 trial, 16 experienced developers took 19% longer with AI on familiar codebases, yet believed they were faster. This is evidence about that setting, not a verdict on today’s tools.
💡 Why it matters Measure completed work, including review and rework, before booking capacity gains. User enthusiasm is a separate metric.
☕ Coffee talk Has finance booked the AI saving before engineering has measured the review time?
3️⃣ Close the project with a change to the next one Research reviews find that after-action debriefs improve performance. Effective reviews examine a specific event using evidence from more than one perspective; a general chat about teamwork is not the same.
💡 Why it matters Compare the plan with what happened. Agree a change for the next delivery and check whether the team actually uses it.
☕ Coffee talk Can anyone name a delivery practice that changed after your last retrospective?