The AI Futures Project — the team behind AI 2027 — published a detailed self-grading of their 2025 predictions in early 2026. This is a unique comparison: the authors themselves assessing how their scenario held up against reality.

We compare their findings with our independent tracker assessments below.

Comparing different assessment methods

The authors’ self-grading is a dated assessment of their scenario. This tracker evaluates individual claims with their original quantities and time windows. Its status counts are not an aggregate pace estimate and should not be translated into a takeoff date.

The comparisons below retain the historical authors’ assessment as context. Follow each dossier for the current independent assessment, its evidence and its timing explanation. A later milestone does not retroactively confirm an earlier deadline.

Prediction-by-Prediction Comparison

Coding & Benchmarks

PredictionAI Futures AssessmentOur TrackerAgreement?
SWE-bench 85% by mid-2025Behind — best comparable score in their grading was 74.5% (Opus 4.1)See linked dossier for current outcome and timingCompare the original metric and date
Coding agents transform development”Fairly accurate” — Claude Code at $500M+ run-rateSee linked dossier for current outcome and timingCompare the original metric and date
METR time horizons0.66×–1.04× of predicted paceSee linked dossier for current outcome and timingCompare the original metric and date
Agents unreliable”Broadly accurate” — coding agents slightly more reliable than expectedSee linked dossier for current outcome and timingCompare the original metric and date

Economic & Revenue

PredictionAI Futures AssessmentOur TrackerAgreement?
OpenAI revenue ($18B by 2025)Slightly ahead (~$20B annualized)The tracker does not score the 2025 target; the linked dossier tests the $45B 2026 revenue milestone.Compare the original metric and date
OpenAI valuation ($500B by Jun 2025)Behind — hit $500B in Oct 2025The tracker does not score the 2025 target; the linked dossier tests the $3T end-2026 valuation milestone.Compare the original metric and date
Infrastructure investmentNot specifically gradedSee linked dossier for current outcome and timingCompare the original metric and date

AI Capabilities & Research

PredictionAI Futures AssessmentOur TrackerAgreement?
First glimpse of AI agentsCorrect — ChatGPT agent (Jul 2025)See linked dossier for current outcome and timingCompare the original metric and date
AI for AI research focusAIs helping with coding, less with other researchSee linked dossier for current outcome and timingCompare the original metric and date
AI R&D upliftBehind pace — estimates revised downwardSee linked dossier for current outcome and timingCompare the original metric and date
Models frequently updatedCorrect — GPT-4o → GPT-5 → GPT-5.1See linked dossier for current outcome and timingCompare the original metric and date

Safety & Security

PredictionAI Futures AssessmentOur TrackerAgreement?
AI good at hacking/bioweaponsOn track — Anthropic upgraded to ASL-3See linked dossier for current outcome and timingCompare the original metric and date
Sycophancy and deceptionNo explicit gradeNot tracked by this siten/a
Model spec / alignment training”Already true at publication”Not tracked by this siten/a

Where They Got It Wrong

PredictionAI Futures AssessmentOur Tracker
Others 3–9 months behind OpenAIRace closer than predicted (0–2 month lead)Behind (85%): the dated ECI comparison shows a smaller public lead than the predicted 3 to 9 months, which contradicts this prediction; internal gaps remain unresolved.
Compute growth (largest training run)Uncertain — no confirmed run much larger than GPT-4.5Not yet testable; public compute evidence remains limited

Key Divergences: Our Tracker vs Their Self-Grading

Where We Agree Most Strongly

  • AI R&D multiplier (1.5×): the tracker rates this Emerging (75%). Meaningful research assistance is documented, but a comparable 1.5× whole-research multiplier was not established within the early-2026 window.
  • Coding transformation is real — the authors call this prediction fairly accurate; the tracker rates it Confirmed (85%)
  • Agents exist but struggle with reliability — the authors call this broadly accurate; our tracker: Confirmed (95%)
  • Leading-lab revenue — our tracker: On Track (75%) for the registered $45B annual-revenue target at end-2026. The authors’ 2025 revenue comparison is a different window and measure, not direct agreement on this target.
  • SWE-bench target — our tracker: Confirmed (90%), with late timing. The authors’ earlier shortfall assessment does not contradict later attainment.

Where Neither of Us Can Say Much Yet


Their timeline updates

These are dated forecasts, not one permanent adjusted timeline. Automated Coder (AC) is the authors’ milestone at which an AI company would rather give up its human software engineers than its AI coding tools; it is not identical to their Superhuman Coder or superintelligence milestones.

UpdateDaniel’s coding-automation medianEli’s coding-automation median
February 12, 2026 grading2029Early 2030s
April 2, 2026 Q1 updateMid-2028Mid-2030
August 16, 2026 Q2.5 updateNovember 2027January 2030

The August values are the all-things-considered medians shown in the update’s combined AC forecast chart, not model-only outputs. April’s values are rounded; the August retrospective chart places Eli’s April median in July 2030.

The grading post estimated 65% pace (February 2026; revised to about 75% in a July 2026 note) for quantitative progress. Its illustrative mid-2028 to mid-2030 takeoff interval was a February extrapolation with compute and labor slowdowns, not a pair of current personal forecast medians. These external pace estimates are not an aggregate score produced by this tracker.


What They’ll Track in 2026

The AI Futures team identified four key metrics to watch:

  1. AI R&D uplift — AI 2027 predicted 1.9× by end 2026. This is the single most important metric for the takeoff timeline.
  2. Revenue & valuations — $55B revenue, $2.5T valuation predicted for 2026. Tests economic traction.
  3. Coding time horizons — ~3 work weeks by end 2026 on central trajectory (via METR measurements).
  4. Other benchmarks — SWE-bench, VideoGameBench, and emerging evaluations.

Our tracker monitors all of these plus governance, geopolitics, and security developments that the self-grading didn’t specifically cover.


Who Is More Aggressive?

AreaMore Aggressive SourceNotes
Overall timelineAI 2027 original (early 2027 takeoff)See the dated February, April and August updates above; coding automation and the subsequent takeoff interval are distinct
Coding progressAI 2027 original timing85% SWE-bench target was reached late; the tracker now rates the outcome Confirmed
Revenue growthNot directly comparableThe authors’ 2025 comparison and the tracker’s $45B end-2026 annual-revenue target differ
AI R&D upliftAI 2027 originalAuthors downgraded their own estimate
Agent deploymentQualitative agreementAuthors: broadly accurate; tracker: Confirmed (95%) for useful but unreliable agents
Lab competition gapAI 2027 originalPredicted 3–9 month gaps; actual is 0–2 months

Bottom Line

The authors’ dated grading and our claim-level dossiers answer different questions. Coding-agent usefulness and unreliable agents have direct support; benchmark attainment can be confirmed while still late, and precise research multipliers remain unresolved. The authors have also revised their forecasts in both directions. These comparisons do not establish one aggregate tracker pace or a uniform delay across the scenario.


Related:

Data source: Grading AI 2027’s 2025 Predictions by Daniel Kokotajlo and Eli Lifland, AI Futures Project (~February 2026)