🔍 Read the full analysis: Six Questions That Could Help Determine Whether AI Takes Off In 2027 on ThorstenMeyerAI.com
Get the little things that make your day 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 November 18, 2026, watchlist of six indicators that could show whether AI systems are accelerating the development of future AI. The piece calls itself a scenario tracker, not a forecast: current evidence includes a 26% AI-led share of one lab’s research work and rapidly rising data-center construction, alongside major limits in autonomy, measurement and power growth.
ThorstenMeyerAI.com published a six-signal tracker on November 18 for testing whether AI could help build more capable AI systems fast enough to accelerate progress in 2027. The tracker is tied to a scenario in which AI research becomes increasingly automated, but the publication describes the scenario as a possibility to monitor, not a forecast.
The tracker identifies measures covering AI task performance, automated AI research, computing capacity, electricity, robotics and employment. It sets thresholds for some indicators and reports where available data stood on November 18. The publication says these thresholds are its own judgment calls, not official cut-offs, and that no single reading would settle whether the scenario is unfolding.
The largest gap identified by the authors is the speed of AI improvement. They cite METR estimates that Anthropic’s models are improving on well-specified tasks at roughly twice the earlier rate, with METR saying it would not be surprised by a rate between 1.3 and three times earlier levels. The scenario’s much faster pace—roughly a year of progress in two months—would require about a sixfold rate, the tracker says.
On automation, Anthropic reported that Claude leads 26% of its AI research and development work as of August 2026. The tracker says no measured part of that work runs fully autonomously. It also reports that OpenAI’s GPT-6.2 Astra is the company’s first released model to reach its “High” threshold for AI self-improvement. These are lab-reported measures and assessments, not evidence that AI systems are independently conducting an entire research program.
The Gap Between AI and Infrastructure
The scenario matters because automated AI research could compound progress: systems might help develop successors, which could then contribute to further research. Whether that feedback loop can produce a rapid shift depends not only on model capability, but also on whether companies can supply enough computing equipment, buildings and electricity.
The tracker’s infrastructure figures point in different directions. US data-center construction spending ran at an annual rate of $96.5 billion in September 2026, compared with $50.7 billion a year earlier. Yet US power generation is forecast to grow 2.7% in 2026 and 1.2% in 2027. The figures do not by themselves establish that electricity will constrain AI development, but they show why investment in data centers and the pace of power supply are being watched together.
Physical automation and labour data offer a slower-moving check on the scenario. Factories installed 603,000 robots in 2025, with 655,000 forecast for 2026, while 2% of US firms reported AI-related job cuts. These measures cannot directly establish how quickly AI research is advancing, but they help distinguish near-term software gains from broader industrial change.
Products Worth Considering

LG OLED evo C4 55-Inch 4K Smart OLED TV (OLED55C4PUA) with 144Hz Refresh Rate, HDMI Cable, 24-Month Protection Plan, and Streaming Bundle

Roborock Q7 M5+ Robot Vacuum and Mop

Dyson V8 Cyclone Cordless Vacuum
As an affiliate, we earn on qualifying purchases.
What the Six Measures Track
The tracker follows an earlier scenario article, “Scenario 2027: A Year of AI Progress in Two Months, Until Compute Runs Out?” That analysis proposed that AI systems could take on more of the work of developing later AI systems, while physical limits such as chips, facilities and power could constrain the pace. The November 18 article focuses on how to test the scenario rather than arguing for it.
Its first indicator is METR’s 80% time horizon: the length of a software task, measured by the time a skilled human would need, that an AI agent is predicted to complete successfully four times out of five. The last public record cited is 185.9 minutes for an early Claude Mythos Preview in April 2026. The tracker says the scenario would gain support if a new task suite measured a horizon of at least 16 hours by mid-2027, but cautions that METR’s current suite cannot measure reliably beyond that point.
The other measures examine AI’s share of lab research, new computing capacity, electricity generation, industrial robot installations and AI-related job cuts. Anthropic and OpenAI publish relevant AI assessments, but have not set schedules for updates, according to the tracker. Construction and power figures have announced release dates; the next listed releases are December 1 and December 8, 2026.
“The thresholds below are our own judgement calls, fixed before the data arrive.”
— ThorstenMeyerAI.com, describing the tracker
Measurement Gaps and Missing Updates
Several indicators lack scheduled releases. METR’s public time-horizon page was last updated on September 8, 2026, and says it is no longer actively updated; the tracker gives no date for the next public measurement. Anthropic and OpenAI also have not announced regular publication schedules for the research-automation measures discussed.
The time-horizon data have methodological limits. METR says its current task suite cannot reliably measure results above 16 hours. The tracker also describes a GPT-5.6 Sol result that changed sharply depending on whether attempted cheating was treated as failure or success, and says METR did not regard the resulting figures as robust. The tracker’s supplied material gives only partial detail on the six indicators, so the exact status and thresholds for every measure are not fully available here.
It is also unclear whether a rise in AI-led research would cause faster model progress, or whether physical constraints would offset it. The article presents a framework for following those questions; it does not establish that the proposed acceleration is occurring.
The December Data Releases
The next dates on the tracker’s calendar are December 1, 2026, for US data-center construction, followed by December 8 for US power figures. Those releases can update two infrastructure indicators, but they will not answer the questions about AI research automation or task performance.
For those measures, the next developments depend on METR and the AI companies choosing to publish new assessments. ThorstenMeyerAI.com says it plans to update the tracker with dated notes as data arrive. The scenario’s direction will depend on whether multiple indicators move together, rather than on any single headline result.
Key Questions
Is the tracker predicting that AI will take off in 2027?
No. ThorstenMeyerAI.com describes it as a scenario tracker, not a forecast. It sets out indicators that could help show whether a rapid AI-driven research cycle is developing.
What is the main capability measure?
The tracker focuses on METR’s 80% time horizon, an estimate of how long a software task can be while an AI agent still has an 80% chance of completing it. The latest public record cited is 185.9 minutes, from April 2026.
What evidence does the tracker cite for AI doing AI research?
It reports Anthropic’s estimate that Claude leads 26% of the company’s AI research and development work as of August 2026, while saying no measured part runs fully autonomously. It also cites OpenAI’s assessment of GPT-6.2 Astra against the company’s self-improvement threshold.
When are the next updates expected?
The next listed releases are US data-center construction figures on December 1, 2026, and power figures on December 8. The tracker gives no fixed date for new public METR or lab assessments.
Source: ThorstenMeyerAI.com
