AI 2027 Tracker

Coding agents vs AI 2027

Observed METR time-horizon measurements since publication, shown alongside the AI 2027 and AI Futures reference points.

80% reliable horizon 50% capability frontier AI 2027 implied pace Today Takeoff windows
METR coding time horizons since AI 2027 publication A log-scale line chart of observed 80 percent and 50 percent METR coding time horizons, with a derived 4-month doubling reference and 2027 original and 2028 to 2030 adjusted takeoff windows shown as background bands. 15m1h4h8h16h1d1w1mo AI 202720262027202820292030 TH1.1 weakens above ~16h Today AI Futures: ~3 work weeks 2.0h 5.9h 18h 30m 1.1h 3.5h Original 2027 window Adjusted 2028-2030 window Derived 4-month doubling, anchored to first 80% point
AI 2027 reference ~4 month doubling

Shown as a derived reference line anchored to the first observed 80% coding time-horizon point.

AI Futures corrected 1.04x

80% coding time horizon compared with their corrected central AI-2027-speed trajectory.

Original displayed graph 0.66x

The same metric compared with the originally displayed, later-corrected trajectory.

Broader quant pace 58-66%

AI Futures' aggregate estimate across quantitative AI 2027 metrics, not just coding horizons.

What the line means

METR time horizon estimates the human task length at which an AI model reaches a given success probability.

Why 80% is primary

80% better matches reliable autonomous work. 50% is useful context, but one success in two is not operational reliability.

Measurement caution

METR cautions that TH1.1 becomes unreliable above roughly 16 hours, so the upper line should not be over-read.

Observed points: METR TH1.1 public data and METR Frontier Risk Report, May 2026. AI 2027 pace line: derived 4-month doubling reference, not an official reproduced curve. End-2026 watchpoint: AI Futures central trajectory marker, shown separately from the anchored reference line.