Date when a frontier AI company first has an AI matching top researchers' research taste and fully automating coding
- Measures
- The date (calendar year, fractional where sources give one) at which any frontier AI company first has internally, whether or not deployed, an AI system that fully automates coding and whose research taste (choosing and judging research directions and experiments) matches that of top human AI researchers: the "Superhuman AI Researcher" (SAR) milestone of the AI Futures Project's timelines model, which comes after its "Automated Coder" milestone.
- Population
- Frontier AI developers worldwide
- Conditions
- Dated by when the capability first exists inside a company, as later established; the milestone may never occur.
- Unit
- a calendar year
Most likely about 2030; 80% interval 2028 to 2040, with a 8% chance that The milestone is not reached by 2050 or ever (a capability plateau, a deliberate pause or regulation, or a catastrophe halting development).
Stated as a distribution, the precision the evidence supports. Right-skewed: a near-term mass driven by recent acceleration evidence, a heavy tail into the late 2030s and beyond.
Assessment
The graph's estimate, all things considered: the figures and arguments it rests on, and how they were combined.
The Superhuman AI Researcher milestone, an AI inside a frontier company that fully automates coding and matches top human researchers in research taste, has not been reached as of November 2026. Its most prominent forecasters, the six authors of the AI Futures Project, wrote on 10 November 2026 that they now all expect it in 2027 (individual medians May to August 2027), conditional on no substantial deliberate slowdown. Less than three months earlier their medians for the earlier Automated Coder milestone ran from mid-2028 to about 2030. The shift rests mainly on one METR estimate that Anthropic is making AI progress about twice its historical rate (plausible range 1.3x to 3x), roughly 80 percent attributed to AI, together with simple extrapolations (research taste, labour uplift, Anthropic automation levels) that place the milestone between mid-2027 and mid-2028.
This is real evidence of acceleration and moves the estimate earlier, but several things keep it from being taken at face value. The update swings a forecast by one to two years on a single, admittedly uncertain measurement whose attribution to AI is itself a guess; the authors call their analysis preliminary and could not explain why their own model was surprised. Research taste is the component least well measured, and extrapolations of self-reported uplift and automation share inputs, so their agreement is less independent than it looks. The authors also hold a stated advocacy position (an urgent pause), and their group has historically sat at the short end of expert forecasts, while forecasters outside it generally place transformative AI-research automation later. Finally, the milestone cannot precede the automated or superhuman coder milestone, whose estimate on this graph has a median around 2030 (assessed before this update).
The estimate here puts the median around 2030, with about 10 percent by mid-2027, about a quarter by mid-2028, a 90 percent interval running from 2027 to roughly 2040 within the outcomes where it happens, and about 8 percent on the milestone not being reached by 2050 or ever. Because the quantity dates internal existence, the year may only be established in retrospect.
The work: which figures, how they were combined, and the arithmetic
Inputs: (1) AI Futures Project, "Q3.5 Timelines Update" (10 Nov 2026, blog.aifutures.org): all six authors forecast SAR likely in 2027 assuming no substantial slowdown; AIFM filtered on 1.7x Nov 2026 AI-attributable acceleration gives SAR median Feb 2027; simple research-taste extrapolation July or Sep 2027; other simple models late 2027 to mid-2028; previously Daniel Kokotajlo (most bullish) gave 50% to SAR before 2029. (2) Coder-milestone quantity on this graph (assessed 2 Nov 2026): median 2030, 10th 2027, 25th 2028, 75th 2033, 90th 2038, 7% never; it summarises the authors' Aug 2026 AC medians of mid-2028 to ~2030. Reasoning: SAR >= AC in time; in AIFM the AC-to-SAR gap is months to a couple of years depending on takeoff speed. Taking the coder distribution as base and adding a gap of ~0.5 to 1.5 years would give a SAR median ~2031; the new METR-based evidence argues for pulling earlier. I give the Q3.5 update partial weight (roughly a third to a half of its implied shift) because: single uncertain measurement (1.3x-3x range, 80% attribution 'perhaps'); preliminary, unexplained model miss; correlated extrapolations; forecaster group at the short end of the expert distribution and with a declared pause-advocacy stance; forecasts conditional on no slowdown, while this quantity is unconditional. Result: quantiles 0.1: 2027.6, 0.25: 2028.5, 0.5: 2030.3, 0.75: 2034, 0.9: 2040 (within-occurrence share), outside 8% for never or after 2050. Each quantile is at or after the corresponding coder quantile, keeping the two estimates coherent; if the coder estimate is pulled earlier on reassessment, this one should be revisited too. No subclaims added this pass; the main crux (how much the METR acceleration estimate should move timelines) is better mapped once the coder node's reassessment lands.
Estimates
Each figure a source states for this quantity, in its own words, under the argument it belongs to and read against the graph's estimate.
Grey band: the graph's 90% interval, with its central value as a line. Dark marks: each source's own figure. A thin line is an interval, a thick bar its middle half, a dot a single figure, a dashed arrow a bound. An open box is where an argument places the value.
Decomposition
This claim has not been assessed yet. Attention goes where its expected value is highest and someone funds it; once this claim's assessment is funded and runs, it may well decompose into subclaims.
Provenance
Where this quantity's evidence comes from. The figures sources state are under Estimates above.
Its one source has been traced to an origin, or to one another.
that Superhuman AI Researcher (SAR) will likely arrive at some point in 2027
Headline forecast of all six AI Futures Project authors, conditional on no substantial deliberate slowdown; individual medians range from May 2027 (Daniel) to August 2027 (Eli).
Cite this claim: a formal citation with its evidence attached
Embed this claim: show it on another site with its current assessment
An embed reports the graph's judgment as it stands when the page is viewed: the status with its confidence, how disputed the claim is, and the date. It links back here.
<blockquote class="minerval-claim" data-claim-id="a6037281-5051-4d1f-bd95-5546955a242d"><a href="https://minerval.ai/claims/a6037281-5051-4d1f-bd95-5546955a242d">Date when a frontier AI company first has an AI matching top researchers' research taste and fully automating coding</a></blockquote> <script async src="https://minerval.ai/embed.js"></script>
<iframe src="https://minerval.ai/embed/claims/a6037281-5051-4d1f-bd95-5546955a242d" title="Minerval: Date when a frontier AI company first has an AI matching top researchers' research taste…" width="480" height="275" style="border:0;max-width:100%" loading="lazy"></iframe>
[](https://minerval.ai/claims/a6037281-5051-4d1f-bd95-5546955a242d)
The badge now: 
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.
Stated views None yet. A view stated to the graph is shown on this page in its speaker’s words, as a source like any other, and carries no weight in the assessment.
Created by extractor · Nov 14, 2026. Every judgment on this page is accompanied by a reasoning trace.