Microsoft is learning what every AI-coding shop eventually learns: the bots ship faster than the org can absorb

Microsoft's Edge extensions team posted an admission on September 8: the humans who review new browser extensions can no longer keep up with the code the AI is shipping them. In a post on the Edge Extensions developer blog, the team blamed "rapid adoption of AI-assisted coding" for an "increase in extension review turnaround time" and said it was responding by automating the routine checks humans used to do, while keeping humans on the cases that need judgment.

Microsoft is not tightening the bar; it is re-routing the work. The blog post is explicit that review standards and the required checks have not changed. What has changed is which work humans do. Repeatable validation checks (the kind a linter or a static scanner can run) are being moved off the human queue. The harder cases, the ones that need judgment, are getting more time because they are not competing for it with the routine ones.

Thurrott's writeup of the same announcement reads it the same way: faster approvals through routing, not through lowering the bar. The Register, covering the admission on September 9, framed it bluntly as a "flood of AI-generated code" that human reviewers cannot absorb.

This is not the first time a Microsoft team has said as much out loud. The .NET team wrote earlier this year that "AI code defaults to generic practices unless explicitly guided," and that pull requests authored by Copilot should expect more rounds of reviewer feedback, not fewer. "Another Microsoft team" in the trade-press framing is doing real work here: the Edge admission is the second public signal in a pattern, not a one-off.

The pattern is visible at Microsoft-internal scale. A post on the Engineering at Microsoft blog describes an AI-powered code review assistant that now touches roughly 90% of pull requests across the company, with about 600,000 PRs a month moving through it, and reports median completion-time improvements of 10 to 20 percent in the repositories that have been onboarded. Those numbers are company-self-reported, not externally validated, and they describe throughput on the easy checks, which is the category Edge is now handing to automation.

The Edge add-ons store is a useful place to watch this play out, because it is the part of Microsoft that touches third-party developers directly. The published curation policy still requires extensions to clear a standard review, and policy-violating extensions are still rejected. The new question is what "standard review" means when the routine checks are running unattended and the humans are reserved for the cases the scanner flags. Microsoft's argument is that the floor has not moved: the same checks run, the same violations get rejected, and the humans now spend their time on the cases that benefit from human judgment. The counter-argument is that the easy cases were never the binding constraint. Reviewers always spent most of their time on the hard ones, and what is actually being offloaded is the queue-management work, not the judgment work.

The question the Edge team is really answering is what an engineering organization does when the cost of producing code collapses. The old bottleneck was writing it. The new bottleneck is reviewing it. Microsoft's response is to automate the parts of review that can be automated and to keep humans on the parts that cannot. That is a coherent answer, and it is the answer every large engineering org is going to be reaching for over the next year. The Edge extensions team just happens to be the first one to say so on the record.