TL;DR

Microsoft’s Edge browser team has acknowledged that AI-assisted development is producing extensions faster than its reviewers can vet them, lengthening turnaround times, and that its answer is to automate parts of the review. It is the second Microsoft team in a matter of weeks to say something similar — the Exchange group recently blamed an AI-generated bug backlog for a delayed cumulative update. For anyone being sold AI coding tools on throughput alone, this is the counter-evidence arriving from the vendor’s own side of the table.

The bottleneck moved, it did not disappear

Because Microsoft vets browser extensions before users can install them, more submissions mean more work — arriving, as The Register points out, at a company that has been shedding staff. An earlier attempt to cope introduced a faster lane for high-quality, high-value submissions. That has since buckled under the volume, with the team conceding that continued growth put its pipeline under further strain and pushed review times up.

The response is automation of the repeatable validation checks plus a streamlined path through the pipeline. Microsoft insists standards are unchanged and that no check has been removed, framing the benefit as consistency on known policy and security violations while freeing reviewers for cases that genuinely need human judgement. It has not said whether that automation is itself AI-based, which is the obvious question.

Alongside this, the team says it can now refresh its “Featured” badge every 15 days, giving developers quicker feedback.

Looking forward

The pattern worth extracting is that AI shifts constraints rather than removing them. Generation gets cheaper; review, verification and assurance do not, and they were already the scarcer resource. Microsoft’s Edge team has essentially reported that gain landing entirely on the wrong side of its own ledger.

UK engineering leaders will recognise the shape from elsewhere: Moody’s has warned that banks now depend on AI they cannot properly test, and the government’s electricity networks review asks for shared testing capability before autonomy expands. Before approving tooling on a productivity case, the useful question is who reviews the additional output, and whether that team has been resourced to match.