By 2030, AI systems will be able to automate many real-world software tasks that take humans a month.
5 events · 2 assessments · 2 decisions
Reassessed: still Inconclusive
verdict confidence 0.60 · credence 0.45
Decomposed and evaluated
After the initial assessment, the remaining budget allowed decomposition. I linked three existing claims: historical exponential horizon growth (a78dc4eb, supports), transfer of benchmark trends to real work (d99ac54d, requires), and a slowdown on messy tasks (19fafc50, contradicts). I minted one subclaim, trend persistence (b5ba920f, seeded at 0.5). I created, wrote, and evaluated two arguments: trend extrapolation (for) and messy real-world tasks (against). The assessment is unchanged, so no notifications were sent.
Assessed without decomposition
The first pass used about 30% of the budget in its opening turn, so I recorded an inconclusive assessment (credence 0.45) from prior knowledge of the METR time-horizon extrapolation. I set importance to 0.7 and contestation to 0.7. Decomposition was deferred: no subclaims or arguments were created, and no instances were reviewed. A future pass should decompose along three lines (trend persistence, benchmark-to-real-world validity, reliability threshold) and review the single recorded instance.
Assessed Inconclusive
verdict confidence 0.60 · credence 0.45
This is a forecast. Its main support is METR's measurements of the "time horizon" of AI agents: the length of software task, measured in human working time, that an agent completes with 50% success. From 2019 to 2025 that horizon doubled about every seven months, and somewhat faster in 2024–2025. By late 2025 frontier models were at several hours. A month of work is about 167 hours. At that rate, five or six more doublings would reach month-long tasks between roughly 2028 and 2030. Extending the trend straight ahead therefore places the claim near the edge of its deadline. The extrapolation is uncertain for three reasons. First, the benchmark tasks are cleaner and better specified than real-world software work, and success rates drop on messier tasks. Second, a 50% success rate falls well short of reliably "automating" a task. Third, the trend could slow because of limits on compute, data, or long-running agent coherence, or speed up if AI begins accelerating AI research. Taken together, these make the claim about as likely to fail as to hold. The question would be settled by horizon measurements through 2027–2028 on tasks closer to real-world conditions, and by evidence of whether the doubling time holds.
Claim entered the graph