KCP 1.2s architecture review¶
April 15, 2026 · LinkedIn
24 reactions · 5 comments · 16,083 views
We cancelled a 45-minute architecture review last week.
Not because the question wasn't important. Because the answer already existed.
The question: "If we change the payment service API contract, what else breaks?"
In any system older than a few years, nobody has the full picture. So the usual move is to find the 4-5 people who each have a piece of it, block out 45 minutes, and spend the first half just assembling what the organisation already knows.
Instead, we ran a query.
synthesis search "payment service API consumers"
1.2 seconds. 14 repos checked. Three downstream services with their contract versions. One SDK. And a batch job that nobody on the current team remembered — running nightly at 03:00, calling the v2 endpoint.
That last one would have been discovered in production.
The meeting still happened. 15 minutes, 3 people instead of 5. We talked about the decision — versioning strategy, migration timeline, who owns the deprecated endpoint. Not about what exists.
This is the shift I keep coming back to: architecture reviews aren't a process problem. They're an infrastructure problem. When organisational knowledge can't be queried, it gets assembled in meetings. That's a reasonable workaround for a structural deficit.
The meeting was never the problem.
Full post on the blog if you want the detail on how KCP manifests make this work.
Discussion¶
Totto ↩: Full blogpost: https://wiki.totto.org/blog/2026/04/15/we-cancelled-a-45-minute-architecture-review-a-kcp-query-answered-it-in-12-seconds/
These lines here hit hard and should be written on stone: These lines here hit hard and should be written on stone
Totto ↩: Matias Luis Lotito Ralli That paragraph took 30 years to write. 😄
The process workaround is so well-established that most organisations have stopped noticing it's a workaround. Change advisory boards, architecture review boards, cross-team syncs — all reasonable responses to the same missing layer.
Appreciate you pulling that out.
Yep, and now imagine this information now available always for all agentic work done by this org. So whenever there is a change, it takes into account payments and other important pieces of information.Can we get to 0 regressions with agentic AI? 👀 : Yep, and now imagine this information now available always for all agentic work done by this org. So whenever there is a change, it takes into account payments and other important pieces of information.Can we get to 0 regressions with agentic AI? 👀
This is a view of how governance can and should be . AI supported and controlled by human risk decision making: This is a view of how governance can and should be . AI supported and controlled by human risk decision making