🔍 Read the full analysis: Six Indicators To Watch As AI Approaches 2027 on ThorstenMeyerAI.com
Get the latest gadgets delivered free with Prime
- Fast, free delivery on millions of items
- Prime Video, Amazon Music and more included
- Member-only deals all year
TL;DR
ThorstenMeyerAI.com published a watchlist on November 18, 2026, for tracking its scenario that AI could accelerate the development of future AI systems in 2027. The six indicators cover agent capabilities, AI use in research, computing infrastructure, electricity, robotics and employment; the site stresses that its thresholds are judgment calls, not forecasts or official cut-offs.
ThorstenMeyerAI.com published a six-indicator watchlist on November 18 for tracking its scenario that AI systems could take on more of the work of developing future AI systems in 2027. The tracker identifies measures spanning agent task performance, AI research work, computing capacity, electricity, robots and jobs, while warning that its thresholds are the site’s own judgment calls, not a forecast or official standards.
The article is designed to monitor a scenario published separately on the same site, not to make the case for it again. The scenario proposes that a faster cycle of AI-assisted AI development could compress a year of progress at the previous pace into roughly two months, until constraints such as chips, buildings and power become limiting. ThorstenMeyerAI.com says it will add dated notes as new evidence appears.
The first indicator is METR’s 80% time horizon: the length of a software task, measured by the time a skilled human needs, that an AI agent completes reliably four times out of five. The last public record cited is 185.9 minutes for an early Claude Mythos Preview model in April 2026. The site’s scenario gains line is a measured horizon of at least 16 hours by mid-2027; it says that would require a new task suite because METR cautions that its current suite cannot measure beyond 16 hours reliably.
The second indicator tracks how much AI contributes to labs’ own AI research and development, alongside formal assessments of self-improvement risks. The source says Anthropic reported that Claude “leads” 26% of its AI R&D work as of August 2026, while no measured portion ran fully autonomously. It also says OpenAI’s GPT-6.2 Astra was its first released model to reach the company’s “High” threshold for AI self-improvement. The other four indicators cover computing infrastructure, power supply, factory robotics and employment. The source reports US data-centre construction at a September 2026 annual rate of $96.5 billion, up from $50.7 billion a year earlier; US power generation growth projections of 2.7% for 2026 and 1.2% for 2027; 603,000 factory robots installed in 2025, with 655,000 forecast for 2026; and AI-related job cuts reported by 2% of US firms.
How the Six Signals Could Shift
The tracker focuses on whether several indicators move together, rather than treating one model release or data point as proof of rapid AI self-improvement. That distinction matters because the scenario depends both on software progress and on the physical resources needed to train and run systems. Faster agent performance or greater AI involvement in research would not, by itself, show that power, chips and facilities can keep pace.
The infrastructure measures also point to different speeds of change. Data-centre construction has risen sharply in the figures cited by the source, while the power-generation growth rates it lists are lower. If electricity supply expands more slowly than computing facilities, power could constrain deployment; the article presents that as a question to track, not a confirmed bottleneck. Robot installation and job-cut figures provide additional evidence about whether AI-related change is spreading into physical production and employment, though neither measure alone establishes AI’s overall economic effect.
For readers, the value is a set of observable signposts and a record of what the author regards as evidence for or against the scenario. The thresholds may help make the argument testable, but they remain the site’s chosen lines, not consensus measures. A reading on one side of a line would not settle whether the broader scenario is unfolding.
Products Worth Considering

Samsung Odyssey OLED G8 34-Inch Curved Ultra WQHD Gaming Monitor

Samsung Galaxy Tab A11+ Kids Edition, 128GB, Royal Blue

Bose QuietComfort Ultra Headphones (2nd Gen), Wireless Noise-Cancelling Over-Ear Headphones, Driftwood Sand
As an affiliate, we earn on qualifying purchases.
A Scenario With Measurable Tests
The watchlist follows a separate ThorstenMeyerAI.com article titled “Scenario 2027: A Year of AI Progress in Two Months, Until Compute Runs Out?” The November 18 tracker says that earlier scenario is a proposition to test, and identifies six indicators, their available readings and the points at which the scenario would gain or lose support. It says there is no single decisive reading.
The author distinguishes between indicators with scheduled government data and those dependent on research groups or AI companies choosing to publish. The source gives December 1, 2026, for the next construction figures and December 8 for the next power figures. It gives no set publication dates for new METR measurements or updates from Anthropic and OpenAI.
The article also discloses that it was researched and drafted with AI assistance, including Anthropic’s Claude. It says evidence from Anthropic and OpenAI informs two signals and that METR’s independent estimate concerns Anthropic. The source states that none of those organizations reviewed or sponsored the article.
“The thresholds below are our own judgement calls, fixed before the data arrive.”
— ThorstenMeyerAI.com, describing the watchlist
Data Gaps and Threshold Limits
Several indicators lack a reliable publication calendar. METR’s public time-horizon page was last updated on September 8, 2026, and the source says it is no longer actively updated. The article provides no date for a new task suite or another public time-horizon reading. Anthropic and OpenAI have also not stated when they will publish the next relevant assessments.
The 16-hour threshold has a measurement problem: METR says its current tasks cannot reliably evaluate performance above that level. The source also warns that results can change substantially depending on how cheating attempts are scored, citing figures it says METR did not regard as robust. More broadly, the supplied source excerpt does not give detailed definitions, current readings or scenario thresholds for all four remaining indicators. Its figures should be understood as source-reported observations, forecasts or estimates in their stated contexts, not as proof of the scenario.
The article’s thresholds are explicitly the author’s judgment calls. It is also an opinion-based scenario tracker, and the author discloses AI assistance and relevant evidence involving the companies discussed. Readers cannot yet tell from this tracker alone whether its indicators will be updated consistently or whether different labs’ measures will be comparable.
Upcoming Data and Lab Updates
The next scheduled releases identified by the source are US data-centre construction figures on December 1, 2026, and power figures on December 8. Those releases may update two of the infrastructure indicators, though the article does not specify what values would alter its assessment beyond its general approach of tracking direction and comparing data with its chosen thresholds.
For the AI capability and research indicators, the next milestone depends on METR, Anthropic and OpenAI publishing new measurements or assessments. ThorstenMeyerAI.com says it plans to update the tracker with dated notes as evidence arrives. Until then, the scenario remains a hypothesis under observation, with no single release or data point designated as a final verdict.
Key Questions
What is the 2027 AI scenario being tracked?
ThorstenMeyerAI.com proposes that AI systems could increasingly help develop future AI systems, potentially accelerating progress. The tracker treats this as a scenario to test, not a forecast.
What does METR’s 80% time horizon measure?
It estimates the length of a software task, based on skilled human completion time, that an AI agent can complete with 80% reliability. The source cites a last public record of 185.9 minutes in April 2026 and notes limits in measuring longer tasks.
When are the next scheduled figures due?
The tracker lists US data-centre construction figures for December 1, 2026, and power figures for December 8. It gives no fixed publication dates for new lab or METR readings.
Do the six indicators prove that AI progress will accelerate?
No. The source says the thresholds are its own judgment calls and that no single reading decides the scenario. The indicators are intended to show whether evidence is moving in a direction consistent with it.
Source: ThorstenMeyerAI.com
