TL;DR
A November 18, 2026, watchlist identifies six indicators for judging a scenario in which AI accelerates the development of future AI systems in 2027. Current evidence is mixed: lab reports show AI taking a growing role in research, while independent task-horizon estimates, infrastructure forecasts, robot deployment and employment data do not yet establish the scenario.
Thorsten Meyer AI published a six-signal tracker on November 18 for a scenario in which AI systems take on more of the work of building future AI systems in 2027. The tracker sets out current readings, sources and thresholds that the publisher says would strengthen or weaken the scenario; it describes a watchlist, not a forecast.
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
The tracker measures AI capability and research automation through three signals: the length of tasks AI can complete with 80% reliability, AI’s share of AI research and development work, and independent estimates of how quickly AI capabilities are improving. METR’s last public record listed in the article is an 80% time horizon of 185.9 minutes for a model from April 2026. The publisher sets 16 hours by mid-2027 as a threshold that would support its scenario. METR’s current task suite, the article says, cannot reliably measure beyond that point.
For research automation, the tracker cites Anthropic’s August 2026 report that Claude “leads” 26% of the company’s AI research and development work, with no measured portion fully autonomous. It also cites OpenAI’s November 17 statement that GPT-6.2 Astra reached its “High” threshold for AI self-improvement. These are company-reported measures with different definitions; the tracker does not treat them as proof that AI research is running autonomously. METR’s October assessment puts improvement on Anthropic’s benchmark index at about 2x, with a stated range of 1.3x to 3x. The tracker says roughly 6x would correspond to a year of progress in two months, the pace implied by its scenario.
The remaining signals concern physical capacity and economic effects. US data-centre construction spending was running at an annual rate of $96.5 billion in September, compared with $50.7 billion a year earlier, according to Census Bureau figures cited by the tracker. The International Federation of Robotics reported 603,000 industrial robot installations in 2025, with a 655,000 forecast for 2026. Separately, the tracker cites a Census Bureau survey in which 2% of US firms reported AI-related job cuts; 18% reported using AI. Each figure has a specific period and source, and the article sets its own thresholds for what would count as support.

What the Six Signals Can Show
The tracker separates claims about software progress from constraints on building and deploying systems. Faster task performance or a larger reported AI role in research would speak to the first part of the scenario. Construction, power and robot data bear on whether physical capacity can keep pace, while employment figures offer one limited measure of effects at firms.
The readings do not all move on the same schedule or measure the same thing. Lab disclosures can describe internal work, while METR’s task evaluations are an independent estimate with stated measurement limits. Construction and employment statistics describe broader economic activity and do not, by themselves, establish that AI caused a change. Taken together, the six indicators can help readers track distinct parts of the scenario without treating any single number as confirmation.
Products Worth Considering

Sonos Era 100 SL Compact Microphone-Free Speaker, Black

LG 27GS60QC-B Ultragear 27-inch Curved Gaming Monitor (QHD, 180Hz, 1ms, 1000R)

Samsung Odyssey OLED G8 34-Inch Curved Ultra WQHD Gaming Monitor
As an affiliate, we earn on qualifying purchases.
How the 2027 Scenario Is Framed
The tracker follows an earlier scenario analysis published by Thorsten Meyer AI on November 18. That analysis argued that AI systems could increasingly help build the next generation of AI, potentially compressing a year of progress into about two months, until limits in chips, buildings and power slow the process. It then described robots, factories and laboratories as a slower physical loop. Those are scenario claims, not established outcomes.
The publisher says the thresholds in the tracker are its own judgment calls, set before future data arrive. It also discloses that AI assisted the research and drafting, including Anthropic’s Claude, and that evidence from Anthropic, OpenAI and METR features in the analysis. It says none of those organizations reviewed or sponsored the article. The tracker is not investment advice.
“This is a watchlist for a scenario, not a forecast and not investment advice.”
— Thorsten Meyer AI, describing the tracker
Limits in the Current Readings
The scenario remains unconfirmed. METR’s listed task horizon is measured in minutes, far short of the tracker’s 16-hour threshold, and the cited task suite cannot reliably measure at that level. METR’s improvement estimate concerns Anthropic and carries a broad range; the source does not establish that the same pace applies across labs or tasks.
The AI research figures also rely on organizations’ own descriptions and may not be directly comparable. The source provides no fixed release schedule for new lab disclosures or METR readings. It does not specify the next power forecast figures beyond noting expected US generation growth of 2.7% in 2026 and 1.2% in 2027, and it does not establish how much of the construction, robot or job data is attributable to AI.
Next Data and Reporting Dates
The tracker lists December 1, 2026, for the next US construction spending figures and December 8 for the next power figures. It gives no announced dates for updates from METR or the AI labs, which publish on their own schedules. The publisher says it will add dated notes as evidence arrives.
Readers can compare each new reading with the tracker’s stated thresholds, while keeping the underlying measure and time period in view. Whether the combined evidence supports the 2027 scenario will depend on later results, including task evaluations beyond current measurement limits and clearer evidence about AI’s role in research and infrastructure demand.
Key Questions
What is the 2027 AI scenario?
It is a scenario in which AI systems increasingly help develop future AI systems, accelerating progress until limits in chips, buildings and power become binding. The tracker presents it as a possibility to monitor, not a forecast.
What would count as a strong signal for faster AI progress?
The publisher identifies an 80% task time horizon of at least 16 hours by mid-2027, a substantial rise in AI-led research work, and independent improvement rates well above 3x as signals that would strengthen its scenario.
Is AI already running AI research autonomously?
The cited Anthropic report says Claude led 26% of the company’s AI research and development work as of August 2026, but the tracker says no measured portion was fully autonomous.
When are the next scheduled readings?
The tracker lists US construction spending figures for December 1, 2026, and power figures for December 8. It gives no scheduled dates for new METR or lab disclosures.
Source: Thorsten Meyer AI
