Independent product studio · Design and build, eight people
Signing off AI-assisted client work without hiring QA
Twelve to fifteen pages a month, half of them drafted with an assistant, one senior designer reading copy between other work. The studio did not want a QA hire; it wanted a step that took twenty minutes and caught the things a client would notice.
- 31
- Findings across the first five deliverables
- 6
- Blocking, all caught before client review
- 20 min
- Added per deliverable
Four done, one to go.
The situation
The studio's exposure was not code quality. It was copy: a value proposition written by an assistant, a case study with a number nobody could source, a page shipped with a placeholder still in the subhead.
The trigger was an invented statistic in a pitch deck. It was plausible, it was in the right format, and it survived three reads because everyone was reading for tone rather than evidence.
What the gate looked like
Every deliverable ran through the same four checks before it entered the client review folder, whatever the discipline: the copy checker on the page text, the document checker on the delivered markdown, the link and anchor pass, and the image check on anything generated.
The reports were saved with the deliverable, not kept separately, so the reviewer opened the page and the findings together.
What the findings actually were
Nine of ten findings were boring and repeatable: unsourced superlatives, placeholder domains in links, two date formats in one document, headings that skipped a rank in an export.
The blocking ones clustered in two places — claims that implied a result the studio could not evidence, and images with no provenance metadata after passing through a design tool. Both are cheap to fix when found at the folder, and expensive when found by the client.
Where a person was still required
No check could decide whether a positioning argument was any good. That remained a design-director judgement, but it started from a report instead of a blank page, and it stopped being interrupted by placeholder hunting.
Other engagements
Series A, fourteen engineers
Reviewing agent-written pull requests without stalling the queue
News and features, thirty staff
Publishing AI images with provenance that survives the CMS
Client names withheld. Figures describe the pattern of these engagements, not audited outcomes.
Your own file
The first report is the useful one.
Run a single deliverable through the console and compare its findings with what your current review already catches.