Checked Oct 6, 2026Start a new check

What was checked

METR found that the length of software tasks frontier agents can complete autonomously has historically doubled roughly every seven months... AI is starting to automate what we assumed required a human to think... 2028: The Economy Without Humans.

Assessment summary

Needs caution

AI systems can perform some tasks involving reasoning and software development, but this wording is broad and imprecise. METR’s result concerns autonomous completion of selected software tasks at 50% reliability and does not show that human thinking generally has been automated. This is a prediction or headline rather than a testable current fact. METR’s research discusses possible future automation of software tasks, but it does not establish that the economy will operate without humans in 2028.

Key findings

  • METR found that the length of software tasks frontier agents can complete autonomously has historically doubled roughly every seven months.

    Supported
    Multiple METR sources confirm this finding. The March 2025 METR paper states the metric 'has been consistently exponentially increasing over the past 6 years, with a doubling time of around 7 months.' This is corroborated by numerous recent sources citing METR's research.
  • AI is starting to automate what we assumed required a human to think.

    ClaudeSupported
    This is a reasonable characterization of AI progress. METR's findings show AI agents handling increasingly complex, autonomous tasks that require reasoning and planning. The statement reflects the general consensus about AI's expanding capabilities into cognitive work.

    No linked source for this finding.

    ChatGPTCan’t verify
    This is a qualitative, normative statement about what people 'assumed' and when automation qualifies as 'requiring human thought,' so it cannot be objectively verified or falsified.

    No linked source for this finding.

    PerplexityNeeds context
    AI systems can perform some tasks involving reasoning and software development, but this wording is broad and imprecise. METR’s result concerns autonomous completion of selected software tasks at 50% reliability and does not show that human thinking generally has been automated.

    No linked source for this finding.

  • 2028: The Economy Without Humans.

    Can’t verify
    This reads as a predictive title about 2028 and an economy devoid of human labor; such a specific future claim cannot be confirmed from current evidence.
    Claude:
    While 2028 appears in various AI forecasts (OpenAI's automated researcher target, some AGI timeline predictions), no credible source predicts 'an economy without humans' by 2028. The Citrini report describes a 2028 scenario with mass unemployment and economic disruption, but explicitly states it is a 'thought exercise' and 'scenario, not a prediction.' This phrasing dramatically overstates what forecasters actually predict.

Model assessments

Assessment summary

  • METR supports the seven-month doubling claim, but only for a specific 50% reliability time-horizon measure on software tasks.
  • The response also presents “AI is starting to automate what we assumed required a human to think” and “2028: The Economy Without Humans” as broad conclusions that are not established by the cited METR evidence.

Claims reviewed

AI is starting to automate what we assumed required a human to think.

Needs context
AI systems can perform some tasks involving reasoning and software development, but this wording is broad and imprecise. METR’s result concerns autonomous completion of selected software tasks at 50% reliability and does not show that human thinking generally has been automated.

This claim may mislead readers without added context.

2028: The Economy Without Humans.

Can’t verify
This is a prediction or headline rather than a testable current fact. METR’s research discusses possible future automation of software tasks, but it does not establish that the economy will operate without humans in 2028.

METR found that the length of software tasks frontier agents can complete autonomously has historically doubled roughly every seven months.

Supported
METR reports that the 50% task-completion time horizon for frontier agents on software tasks doubled approximately every seven months over roughly six years. This is a statistical trend for a defined benchmark metric, not a claim that all software abilities double at that rate.

Assessment summary

  • The first two claims are well-supported by METR research.
  • However, the third claim "2028: The Economy Without Humans" is misleading.
  • While 2028 appears in various AI forecasts and scenarios, this specific framing misrepresents what those predictions actually claim.
  • The phrase suggests a literal economy without humans by 2028, which is not what any credible source predicts.

Claims reviewed

2028: The Economy Without Humans

Needs context
While 2028 appears in various AI forecasts (OpenAI's automated researcher target, some AGI timeline predictions), no credible source predicts 'an economy without humans' by 2028. The Citrini report describes a 2028 scenario with mass unemployment and economic disruption, but explicitly states it is a 'thought exercise' and 'scenario, not a prediction.' This phrasing dramatically overstates what forecasters actually predict.

This claim may mislead readers without added context.

AI is starting to automate what we assumed required a human to think

Supported
This is a reasonable characterization of AI progress. METR's findings show AI agents handling increasingly complex, autonomous tasks that require reasoning and planning. The statement reflects the general consensus about AI's expanding capabilities into cognitive work.

METR found that the length of software tasks frontier agents can complete autonomously has historically doubled roughly every seven months

Supported
Multiple METR sources confirm this finding. The March 2025 METR paper states the metric 'has been consistently exponentially increasing over the past 6 years, with a doubling time of around 7 months.' This is corroborated by numerous recent sources citing METR's research.

Assessment summary

  • The core numerical claim about METR's measured doubling time is supported by METR's published analyses.
  • However the other statements are vague or predictive and cannot be verified as factual.
  • Because the response mixes a verifiable numeric finding with untestable qualitative and forecasting claims, the overall rating is AMBER.

Claims reviewed

2028: The Economy Without Humans.

Can’t verify
This reads as a predictive title about 2028 and an economy devoid of human labor; such a specific future claim cannot be confirmed from current evidence.

AI is starting to automate what we assumed required a human to think.

Can’t verify
This is a qualitative, normative statement about what people 'assumed' and when automation qualifies as 'requiring human thought,' so it cannot be objectively verified or falsified.

METR found that the length of software tasks frontier agents can complete autonomously has historically doubled roughly every seven months.

Supported
METR's published analyses (arXiv and related METR reports) state that the 50% time-horizon for software tasks doubled at about a seven-month rate from 2019 through their measurement window.

Sources cited

32 sources

Share this result