AI Engineering8 min read
Write the evaluation set before you pick the model
Every AI project that went sideways on us had the same root cause: nobody could say what 'good' meant until after the demo.
Delivery
The four-hour overlap rule
Distributed delivery does not fail on skill. It fails on the six hours a day where nobody can answer a question.
- Published
- 9 May 2026
- min read
- 6 min read
- Written by
- The Attatva engineering team
All insights
The failure mode
Distributed engagements rarely fail because the remote team cannot do the work. They fail on question latency. An engineer blocked at 10:00 who cannot get an answer until 16:00 has lost the day, and has usually spent it building something on top of a guess.
Why four hours
Kathmandu is UTC+05:45, which puts us four to five hours ahead of CET and two hours behind SGT. We guarantee four hours of live overlap because below that, the daily standup consumes the window and nothing is left for the ad-hoc question that actually unblocks someone. Above four, the marginal return drops sharply.
The overlap is not the fix, it is the fallback
The real fix is writing things down so the question does not need asking. Decisions in the repository, not in a call. A design note before the branch, not after the review. Teams that rely on the overlap window to transmit context have simply moved the bottleneck rather than removed it.
Handover starts in week one
Documentation written in the last month of an engagement is written by people who have already mentally left, for people they will never meet. We write it as we go and we test it the same way you test a runbook: someone who was not there tries to follow it while the author stays silent.
Read next
Tell us what's breaking.
Send the problem, not a spec. We'll come back within one business day with an honest read on whether we're the right team — and who to talk to if we're not.
attatvasystems@gmail.com · Replies within 1 business day, Kathmandu time (UTC+05:45)