- date:
- org:
- Infrakit · Head of product team
- category:
- Release quality
- title:
Make the release call from test evidence instead of hiring QA
- context:
- With AI-assisted building, more changes reach a release faster. Leadership wanted someone to own the overview of release quality. The obvious answer was a dedicated QA hire.
- decision:
- Proposed that I make the go/no-go call for releases, based on test data (which tests exist and what they show) plus my view of customer cases and how fragile they are, with end-to-end and regression tests part of the developers' definition of done.
- hypothesis:
- Releases that hurt fragile customers become rare. Missing test evidence is visible before a release, not after it. The call does not turn into a queue waiting on me.
- confidence:
- 60%
- review_date:
- reversibility:
- two-way
- related:
- D-015
- tags:
- proposal, release-quality, testing, ai-era
When building gets faster, releases get riskier unless something else changes. More changes ship, and each one has less human attention on it. Leadership asked for one person to hold the overview of release quality, and the standard move is to hire QA for that.
I think the overview needs two things a new hire would not have. The first is evidence: which tests exist for what is shipping, and what they show. The second is context: which customers are in a fragile spot right now, which projects are mid-delivery, and who would get hurt by a regression this week. I already carry the second one. So the proposal is that developers produce the first one and I make the call.
How it works
Tests are part of the definition of done. Not only unit tests, but end-to-end and regression tests for what was built. Bigger changes also come with a short readiness note. All of it lands in one view before a release. Anything without test evidence shows up as a list of open items, so I don’t have to remember to ask. I look at that list together with what I know about customers and say go or no-go.
Alternatives considered
- Hire a dedicated QA person. Real capacity, but they would start without the customer context, and it makes quality someone else’s job instead of the developers’.
- Rely on developers’ own readiness reports. Cheap, but without tests behind it this turns into a pile of “it’s fine”.
Cost and trade-off
I become a checkpoint for every release, on top of everything else. If I am slow, releases wait. It also only works if developers really write the tests. If they don’t, the view is empty and the call is a guess. Building the gate so missing evidence is visible automatically takes some upfront work.
Status
Proposed Nov 2026, not yet decided or rolled out. Two-way: if the call becomes a bottleneck, hiring QA later is still on the table, and the tests written by then are useful either way.
What would change my mind
If releases start waiting on me, or if the test evidence stays thin after the first months, the model is not working and a QA hire comes back as the answer.
Lessons so far
The question was never really “who does QA”. It was who has the context to judge risk, and what data they need to do it without being in every detail.
What I can’t share
Customer cases and which ones are considered fragile, test coverage numbers, release cadence, tooling details and the internal discussion around the approval. Colleagues will recognise this as our release process.