Meta's AI now writes tests no human checks. On purpose.
Meta says "traditional testing can’t keep pace" with hundreds of millions of lines of code. So its AI writes a test for each risky change, after the change, "designed to fail", and engineers only "confirm whether an observed behavioral change was intended." In the first half of 2026 it helped catch 74 serious privacy bugs before they landed.
Source: "Catching Bugs Before They Land: JiTTesting’s Human-on-the-Loop Approach at Full Scale", Engineering at Meta, Oct 29, 2026 (Sreeja Thummala, Sophie Zeng, Mark Harman, Weini Liu, Eric Liu) https://engineering.fb.com/2026/10/29/security/catching-bugs-before-they-land-jittestings-human-on-the-loop-approach-at-full-scale/ Also: Mark Harman, "The Death of Traditional Testing" (Engineering at Meta, Feb 11, 2026); "Just-in-Time Catching Test Generation at Meta" (arXiv 2601.22832, Jan 30, 2026).
The code stream, the review seat, the timeline, the bug, the judges, the loop and the question card are dramatizations; every quote and number on screen is from the sources.
Would you trust a test you never read?
Transcript
Meta's AI now writes tests, no human checks, on purpose. It's in Meta's own article, "Traditional testing can't keep pace." The old way, you write the code, then tests for what it should do. They pass and stay forever. But they were written before your next bug existed, so they keep passing and it ships. Meta flips it. After your change, an AI writes tests just for that change, built to fail it, then throws them away. But in Meta's first trial, only one alarm in five was a real bug. So an AI jury and a strict rulebook throw out the false alarms first. But the human isn't gone. They're on the loop, setting the rules, judging the results. You don't get test code, you get one question. This used to be true, now it's false. Did you mean that? 589 engineers answered. 60% of the alarms were real, they say. So in 6 months, it helped catch 74 serious privacy bugs before they landed. The tests are the machine's job. The question is yours…
