Horizon / Alternative assumptions
What if our assumptions change?
Compare three outlooks for each future. These original scenario windows are possibilities to explore. Our current predictions have their own target dates and tests.
Snapshot 22 Sept 2026 · UTC
Time counts down. Evidence can move the target. Days remaining to each editorial scenario’s boundary, not a promised arrival date. A range counts to its opening year, then the end of its final year. “By” includes the final year. How the clocks work ↓
Published evidence reviews refresh while this page is open.
The CEO question · Agents at work
Use 10× as a design challenge. Measure the gain before claiming it.
Know if it worksFor each 100 cases, compare completed work, exception handling time, recovery effort and customer outcomes with the baseline.
Bars show the stated date windows. Editorial scenarios, not probabilities. Date assumptions from 22 Apr 2026.
What to watch · Agents at work
Repeated production workflows that report completion, human intervention, recovery, cost and failures, including changed tools and missing context.
Explore the evidenceThe evidence behind the question
Sources checked
Our assessment. Task completion is measurable, but reliable ownership of a workplace process is a broader claim. The dates remain illustrative, with no new adoption probability assigned.
01 / Page updated 8 May 2026
METR estimates task difficulty at specified success rates, using human completion time as the reference.
The limitThese are mainly well-specified software tasks. A time horizon is not continuous autonomous runtime or proof that a whole job can be automated.
02 / 24 July 2026
The research note compares performance measures that account for time, cost and the human baseline.
The limitIt analyses measurement choices. It does not establish widespread enterprise adoption.
A view that can change its mind
Agents at work · pragmatic
Every recorded review, including decisions to keep the same window. Changing the outlook above compares assumptions. Evidence revisions appear in this record.
Previous wording: 2029-2031
Reviewed wording: 2029-2031
Threshold at this review: AI systems that autonomously execute multi-step knowledge work across tools, queues and approval boundaries, owning outcomes end-to-end rather than assisting a human operator.
Task completion is measurable, but reliable ownership of a workplace process is a broader claim. The dates remain illustrative, with no new adoption probability assigned.
METR estimates task difficulty at specified success rates, using human completion time as the reference.
Limit: These are mainly well-specified software tasks. A time horizon is not continuous autonomous runtime or proof that a whole job can be automated.
The research note compares performance measures that account for time, cost and the human baseline.
Limit: It analyses measurement choices. It does not establish widespread enterprise adoption.
Threshold at this review: AI systems that autonomously execute multi-step knowledge work across tools, queues and approval boundaries, owning outcomes end-to-end rather than assisting a human operator.
The original scenario wording, preserved before adding the live clocks. This is a dated editorial baseline.
Read the preserved baseline ↗The question behind the timeline
Start with what you mean by general intelligence. The definition changes the evidence you need.
Our reading of the sources, checked . Explore the criteria before drawing a conclusion.
Google DeepMind’s Levels of AGI framework treats breadth and depth of performance as distinct dimensions. Autonomy is a related deployment question.
We would look for broad, independently measured performance, with unfamiliar tasks and the comparison group specified.
Calling something an early level of AGI is a different claim from demonstrating expert performance across most domains.
Read Levels of AGI ↗OpenAI’s Charter connects AGI with highly autonomous systems that exceed human performance in most economically valuable work.
We would look for repeatable outcomes across a representative range of real work, including the assistance and costs involved.
An impressive result in one profession, a selected task or a benchmark does not settle a claim about most valuable work.
Read the OpenAI Charter ↗ARC-AGI-3 tests exploration and adaptation in interactive environments. It offers a concrete way to examine part of the generalisation question.
We would look for repeatable adaptation with limited prior knowledge and a published account of the tools, attempts and resources used.
This is a test of adaptive reasoning, rather than a complete definition or certificate of AGI.
Explore ARC-AGI-3 ↗Showing Breadth of ability.
01 / What we can observe
Dated observations, the original sources and what each one actually establishes.
Read the current record ↗02 / How we got here
From earlier breakthroughs to the tools and ideas that shaped today’s AI.
Explore the history ↗03 / When our view changes
Original claims, revisions and withdrawals. A useful outlook has room to change its mind.
Inspect the decisions ↗These windows are editorial scenarios. They do not have calibrated probabilities. The current assessment and dated decisions for each future appear in its evidence and clock history above.
The original arrival windows date from 22 Apr 2026. Source publication dates and our review dates are shown separately. A new site build does not renew either.
The large number shows whole days remaining, with hours, minutes and seconds underneath. Less than one day is shown as <1 day. The counters use UTC and your device’s clock. A year range counts to 1 January of its opening year, then to the end of its final year. A “by” claim counts to the end of that year. These are display conventions for broad scenarios, not predictions of an exact day. Seconds show the passage of time, not forecast precision.
When a boundary passes, the display says the window or deadline has elapsed. It never declares that a scenario has arrived. Open-ended wording has no numeric countdown. Pause stops the counters; resume catches up with the current time.
Targets change only when we publish a recorded evidence review. The open page checks for published updates every five minutes and when you return to it. It retains the last verified review if that check fails. This does not mean new research is being assessed continuously. Earlier entries remain visible when we change a window, revise its scope or keep it unchanged.
Optimistic, pragmatic and sceptical describe different assumptions. “Pragmatic” is a scenario label, not a claim of expert consensus or a statistically most likely outcome. The education scenarios include both institutional change and improvement within existing institutions.
We link primary research and publisher documentation, state their limits, and describe our own interpretation as an assessment. The separate, testable forecasts retain their September 2026 review and resolution criteria.
Inspect the clock history ↗ · Read the published clock data ↗ · Read the original scenario wording ↗