AI systems will attain superhuman coding ability by 2027.
Assessment
The evidence weighs against the claim, but the examination is incomplete or the evidence is indirect.
The claim is a forecast. Its best-known version comes from the "AI 2027" scenario, which named a "superhuman coder" milestone: an AI system able to do the work of a leading AI company's best engineers far faster and more cheaply. The successor "Automated Coder" milestone is the point at which the leading AI company would rather fire its human software engineers than give up AI for coding. On weaker readings, such as beating most humans at competitive programming, the claim is arguably already true. The live dispute is about the demanding reading.
The forecasters closest to the claim disagree among themselves. In their August 2026 update, the AI Futures Project put Daniel Kokotajlo's median for Automated Coder at about November 2027, with a 10th percentile of January 2027. Eli Lifland's median was around 2030. The authors judged that real progress was running at about 70 to 90 percent of AI 2027's pace, and that at 75 percent of that pace the milestone would arrive around mid-2027. Their forecasts are explicitly conditional on progress going as fast as is technically feasible. Much of the remaining spread comes from how much AI currently speeds up engineers. Independent measurements (METR, about 1.04 to 1.2 times) are far lower than AI companies' internal surveys (about 4 times at Anthropic).
On balance, arrival by the end of 2027 is a serious possibility but less likely than not. Even the most aggressive major forecaster puts it at roughly a coin flip. Other forecasters are later. The conditional framing and reliance on self-reported uplift probably bias these numbers early. Agentic coding at frontier labs over the next year will largely settle the matter: either AI systems take over the bulk of engineering work, or measured productivity gains stay modest.
Full reasoning: the evidence and decisions behind this verdict
Prior model: the demanding reading (automating top engineers' work) looked unlikely by the end of 2027, given modest measured productivity gains and a long tail of open-ended tasks. I had believed the AI 2027 authors had pushed their timelines back to around the end of the decade. Evidence moved this. The AI Futures Project's "Q2.5 2026 Timelines Update" (16 Aug 2026, www.lesswrong.com/posts/ZPSsmRH5oMwLPXys4/q2-5-2026-timelines-update-uplift-and-revenue) says timelines "got slightly shorter". It anchors the Automated Coder forecast on coding uplift (Kokotajlo's medians: present uplift 2x, a 5-month doubling time for uplift minus 1, and 20x uplift needed for AC). It grades reality at 70 to 90 percent of AI 2027's pace. A secondary summary (x.com/rryssf/status/2100240736960799059) reports Kokotajlo's AC median of November 2027 (10th percentile January 2027, 90th percentile April 2030) and Lifland's of January 2030. The same summary reports an audit-driven correction that lowered P(superhuman coder by end of 2027) under the older model with Lifland's parameters from about 55 percent to about 30 percent.
So I revised my credence from 0.15 to 0.25 but kept "doubtful". Reasons: (a) only the most aggressive forecaster is near 50 percent; (b) the forecasts assume progress as fast as is technically feasible, with no slowdowns; (c) the present-uplift input rests heavily on employee self-reports (Anthropic survey, 4x) against the measured METR figure of 1.04 to 1.2x; (d) the uplift extrapolation assumes the exponential holds. Remaining uncertainties: the operational definition, and whether "by 2027" means by the end of 2027 (assumed here). The Gary Marcus instance argues that the AI 2027 appendix shows only that superhuman coding "could", not "will", arrive in 2027. That is consistent with this verdict.
Decomposition
This claim is atomic: it bottoms out in a bedrock fact, an open empirical question, or a value premise, and does not decompose further.
Provenance
Where this claim has been said, linked to its canonical form.
At most it shows that superhuman coding — just one of many prerequisites for the overall scenario — could come in 2027, not, by any stretch, that it will.
Discussing the AI 2027 timelines-forecast appendix.
Assessment history
0 status changes over 2 assessments. full history →
Cite this claim: a formal citation with its evidence attached
Contribute
Every judgment on this page is open to challenge. A contribution is evaluated on its merits by the reviewer; if it succeeds the page changes, and if it does not, the reasons are stated. Either way the exchange becomes part of the claim’s public record.
The attention this claim received was paid for by a funded mandate. Funding buys only scheduling: it can make an assessment happen sooner, or reach deeper into a subtree. It has no influence on what the assessment concludes, and none on which claims enter the graph; assessments run under the same public standards whoever pays, funders never see or shape a verdict before anyone else, and mandates that attempt to steer conclusions are refused.
Created by extractor · Nov 1, 2026. Every judgment on this page is accompanied by a reasoning trace.