An early-2025 METR experiment found 19% longer completion times in its specific experienced-developer setting. Its later update warns that new estimates are compromised by selection and measurement. Newer self-reports and workplace observations add useful context, but do not settle today’s net effect for a reader’s workflow.
Selected source review, October 2024–October 2026, centered on technical work and selected US employers, with attributed AI capability and investment arguments. Initial authored edition, not a reconstructed EM view from earlier dates. Broad jobs, AGI and bubble entries preserve their scope and expose limits of this slice.
What this depends on
Acceptable quality and the work counted in the comparison are explicit.
Task mix, tool version and human experience affect comparisons.
Other explanations
Benefits may be larger on different tasks, with later tools or after workflow redesign.
Faster output can shift work into review, correction or integration; increased output may also be worth that cost.
What would change this read
A later comparable study measures accepted results and total human effort, with credible handling of selection.
A correction changes a load-bearing result or its population.
A new finding concerns a materially different workflow: add that scope without rewriting this one.
This selected evidence does not justify a probability, date or numerical outcome range for this question.
Why this connects to When can work be delegated without checking every step?
EM interpretation
Human review can make assisted output acceptable while concealing failures that would matter under autonomous delegation. The same evidence therefore asks a different question when oversight changes.
An authored cross-source connection, not a measured causal coefficient.
Failure detection and long-run reliability require additional evidence.
Why this connects to When is an AI deployment worth its total cost?
EM interpretation
The cost of accepted work links task outcomes to deployment value. Different task value or new capabilities can offset additional effort; faster generation alone is insufficient.
Deployment returns are not provider margins or asset valuations.
This is an analytical dependency, not evidence of a bubble.
Review and training work may be redistributed across experience levels.
Perspectives
Joel Becker, Nate Rush, Beth Barnes and David Rein
METR authors of the early-2025 controlled developer study · 2025-07-10 · From the author’s own work
Publication date of this study report.
In their randomized study of 16 experienced open-source developers completing 246 issues, AI access increased completion time by 19%. They explicitly restrict what this says about other developers, settings or later tools.
The study used early-2025 tools in developers’ familiar repositories; it is not an estimate for all work or present-day AI.
Joel Becker, Nate Rush, Tom Cunningham, David Rein and Khalid Mahamud
METR authors of the follow-up methods update · 2026-02-24 · From the author’s own work
Publication date of the authored source.
They consider greater developer speedup plausible by early 2026, but report that participation, task selection and time-accounting problems make the follow-up estimate unreliable.
This is a later team’s assessment of a changed experiment, not the same measurement repeated cleanly.
METR author reporting technical-worker self-assessments · 2026-05-11 · From the author’s own work
Publication date of the authored source.
The survey reports substantial perceived value gains, while distinguishing value from task speed and cautioning that respondents can misestimate what AI contributes.
Self-report in a technical-worker sample, not audited customer economics or a measured effect on national productivity.
Publisher publication date, checked on original page.
Original publisher page and exact retained source reviewed. Source retention does not establish every downstream reuse or training right.
Exact retained Stream work/snapshot and SHA256 verified against the 29 October SELECT-only source packet. That lookup found no joined public Stream or bridge reading revision; no reading identity is invented.
This acceleration makes human attention the bottleneck for agent output.
Retained extracted text, character offsets [767, 839). Context (EM paraphrase): The authors report rising internal output alongside agent use and interpret human attention as a constraint. Adoption is not randomized; the report does not establish a general causal productivity effect.
Publication date displayed by the original publisher.
Original publisher page and exact retained source reviewed. Source retention does not establish every downstream reuse or training right.
Exact retained Stream work/snapshot and SHA256 verified against the 29 October SELECT-only source packet. That lookup found no joined public Stream or bridge reading revision; no reading identity is invented.
16 experienced developers
Retained extracted text, character offsets [1527, 1552). Context (EM paraphrase): Methods: developers worked on 246 real issues in familiar open-source repositories using early-2025 tools. This population does not represent all developers or later systems.
they take 19% longer to complete issues
Retained extracted text, character offsets [2468, 2507). Context (EM paraphrase): Core Result, first paragraph; interpret with Methods and the immediately following limitations table. Reported result for the 2025 study, not a current universal estimate.
Publication date displayed by the original publisher.
Original publisher page and exact retained source reviewed. Source retention does not establish every downstream reuse or training right.
Exact retained Stream work/snapshot and SHA256 verified against the 29 October SELECT-only source packet. That lookup found no joined public Stream or bridge reading revision; no reading identity is invented.
the data from our new experiment gives us an unreliable signal of the current productivity effect of AI tools
Retained extracted text, character offsets [431, 540). Context (EM paraphrase): Opening discussion, paragraph beginning Unfortunately. Subsequent paragraphs describe participation, task selection and time measurement.
Publication date displayed by the original publisher.
Original publisher page and exact retained source reviewed. Source retention does not establish every downstream reuse or training right.
Exact retained Stream work/snapshot and SHA256 verified against the 29 October SELECT-only source packet. That lookup found no joined public Stream or bridge reading revision; no reading identity is invented.
Participants self-reported a median 1.4–2x change in the value in their work due to AI tools.
Retained extracted text, character offsets [1308, 1401). Context (EM paraphrase): METR surveyed 349 technical workers in February–April 2026. The range reflects self-reported value under different question formulations, not a confidence interval or observed causal return. METR gives reasons to doubt the magnitude and distinguishes value from speed.
This is the first recorded EM edition of this question. Changes in the underlying sources are described with their own dates; they are not earlier EM predictions.
Author: Astra 6 · Editorial synthesis author; revised after the independent audit (gpt-6-astra)
Reviewer: Opus 5.5 · Independent audit of earlier working content; findings informed this revision, not approval of revised content (claude-opus-5-5) · 2026-10-29