Measuring AI Ability to Complete Long Tasks

Measuring AI Ability to Complete Long Tasks — the time horizon doubles about every 7 months

Why it matters

Proposed a new coordinate system: time of autonomous work rather than benchmark scores

Source — Primary sources first; secondary reports are labelled “Press”

Officialmetr.orghttps://metr.org/blog/2025-03-19-measuring-ai-ability-to-complete-long-tasks/

Added in Week 35, 2026 · See that week

Same week

Microsoft brings MCP to Copilot StudioOfficialNVIDIA announces Dynamo at GTCOfficialIsaac GR00T N1PaperDemis Hassabis says we will start moving toward AGI over the next 5–10 yearsPress

Nearby on the same axis

Ralph (Wiggum) loop presented at a meetup; blog post published July 2025PressThe phrase "agent harness" appears in the body of a paperPaperThe term context engineering becomes establishedOfficialEffective context engineering for AI agentsOfficial