🔍 Read the full analysis: Six Indicators That May Show Whether AI Gains Momentum In 2027 on ThorstenMeyerAI.com
Prime made for students and young adults
- Fast, free delivery for dorm and study essentials
- Prime Video and Amazon Music included
- Member-only deals
TL;DR
ThorstenMeyerAI.com has published a six-indicator watchlist for testing a scenario in which AI helps accelerate the development of future AI systems in 2027. The article labels its thresholds as judgment calls, not forecasts; current evidence is mixed, and several key measurements have no scheduled release.
ThorstenMeyerAI.com published a six-indicator watchlist on November 18 for assessing whether AI progress may accelerate sharply in 2027, as AI systems take on more work involved in developing future systems. The tracker is explicitly a scenario monitor, not a forecast, and says its thresholds are the authors’ judgment calls rather than official benchmarks.
The tracker follows a scenario the site published earlier: AI systems increasingly help build and improve subsequent AI systems, potentially compressing a year of progress at the previous pace into about two months. The scenario’s author says that cycle could eventually run into limits involving chips, buildings and electricity, after which expansion would depend more on physical infrastructure such as robots, factories and laboratories. Those are proposed outcomes, not established predictions.
Its first two indicators concern capability and AI’s role in AI research. The article uses METR’s 80% time horizon—the duration of a task, measured by human expert completion time, at which an AI agent is predicted to succeed four times out of five—and lab disclosures about AI-assisted research. The tracker says the last public METR record was 185.9 minutes for an early Claude Mythos Preview in April 2026. Its scenario gains line is a measured horizon of at least 16 hours by mid-2027, but METR says its current task suite cannot reliably measure above that range.
For research automation, the article reports Anthropic’s August 2026 estimate that Claude “leads” 26% of the company’s AI research and development work, while no measured portion runs fully autonomously. It also points to OpenAI’s GPT-6.2 Astra as the company’s first released model to reach the “High” threshold for AI self-improvement. The remaining indicators concern data-centre construction, electricity generation, industrial robot installations and employment effects. The supplied tracker excerpt gives current figures for several of these, but does not include full descriptions of all six indicators or every threshold.
Testing AI Acceleration Against Data
The watchlist turns a broad claim about rapid AI progress into observable indicators that readers can compare over time. If capability evaluations, lab reports and infrastructure measures all move in the scenario’s predicted direction, that would offer more support than any one announcement or benchmark alone. Conversely, stalled measurements or physical constraints could weaken the case for rapid compounding.
The tracker also draws attention to a potential mismatch between software speed and the pace of physical expansion. It reports US data-centre construction at a $96.5 billion annual rate in September 2026, compared with $50.7 billion a year earlier, while expected US power-generation growth is 2.7% in 2026 and 1.2% in 2027. These are figures cited by the source, not proof that electricity supply will limit AI development. They help frame why construction and power are included alongside model capability.
For readers, the distinction between a testable scenario and a forecast matters. The author’s thresholds are not validated cut-offs, and a reading crossing one line would not by itself show that AI is building better AI at a self-sustaining pace. The tracker says the scenario should be judged by whether multiple signals move together.
Popular right now

Breville Barista Express Impress Espresso Machine with Grinder, Brushed Stainless Steel

De'Longhi Magnifica Evo Fully Automatic Espresso Machine with Milk Frother, Silver

Philips 5500 Series Fully Automatic Espresso Machine with LatteGo Milk System
As an affiliate, we earn on qualifying purchases.
The Tracker’s Baseline Evidence
The article identifies substantial uncertainty in its most direct capability measure. METR’s public results page, according to the tracker, was last changed on September 8, 2026, and is no longer actively updated. The last public record cited is 185.9 minutes; a METR report separately judged that labs’ internal frontier in February–March 2026 was likely at an 80% time horizon of at least three but under four hours. The tracker says no date has been announced for another public measurement.
For automation, the source describes Anthropic’s R&D Automation Index as a prototype using a scale developed by Epoch AI, from no AI involvement to full autonomy. It says OpenAI publishes self-improvement assessments in system preparedness materials. Neither organization had stated a schedule for future readings, according to the article. These measures are not interchangeable: one concerns how work is divided, while another evaluates formal risk thresholds.
Infrastructure and labor provide a different kind of evidence. The tracker reports that factories installed 603,000 industrial robots in 2025, with a 2026 forecast of 655,000, and that 2% of US firms reported AI-related job cuts. It characterizes robot adoption and labor changes as moving at industrial speed. The article identifies December 1 and December 8, 2026, as the next announced dates for construction and power figures; AI-specific updates depend on when labs and METR publish them.
“This is a watchlist for a scenario, not a forecast and not investment advice.”
— ThorstenMeyerAI.com tracker
Key Measures Still Have Gaps
The main uncertainty is whether the indicators can be measured consistently and often enough to test the scenario. METR has not set a date for another public result, its existing suite is not reliable above 16 hours, and the tracker says lab automation disclosures have no announced schedule. A long task horizon would also need careful interpretation: the source notes that attempted cheating can change results substantially, and METR did not consider the cited figures robust measurements.
The evidence presented is also incomplete in the supplied article material: it describes the tracker’s headline and baseline figures, but the available text cuts off during its second signal and does not provide the full wording, thresholds or publication details for all six indicators. The source’s construction and power figures are reported with forthcoming release dates, but the exact implications for AI capacity remain uncertain. The author also discloses that the article was researched and drafted with AI assistance, including Anthropic’s Claude, and says the companies and METR central to parts of the analysis did not review or sponsor it.
Finally, the scenario’s “year of progress in two months” framing is a hypothesis. The tracker sets a rough comparison—the 16-hour threshold is about five times the last public record—but does not establish that crossing it would produce that rate of overall progress. Its thresholds are the author’s own judgment calls, and no single reading is presented as decisive.
December Data and Later Updates
The next scheduled releases identified by the tracker are US data-centre construction figures on December 1, 2026, and power-generation figures on December 8, 2026. The article says it will add dated notes as information arrives. No publication date is given for the next METR time-horizon result or for new AI research-automation readings from Anthropic or OpenAI.
As those figures emerge, the useful test will be whether capability, research automation and physical capacity show a consistent pattern rather than whether one headline number crosses a threshold. The tracker’s 2027 scenario remains open: its own stated approach is to watch the measures, note their limitations and revise the assessment as new evidence becomes available.
Key Questions
Is the article forecasting that AI progress will accelerate in 2027?
No. It calls itself a scenario tracker, not a forecast, and says its thresholds are judgment calls designed to monitor whether the scenario is developing.
What is the tracker’s main capability threshold?
It says the scenario gains support if METR or another suitable evaluation records an 80% task time horizon of at least 16 hours by mid-2027. The source cautions that METR’s current task suite cannot measure reliably beyond that range.
What does Anthropic’s 26% figure describe?
According to the tracker, Anthropic said that in August 2026 Claude “leads” 26% of the company’s AI research and development work. The article says no measured part of that work ran fully autonomously.
When are the next figures expected?
The tracker names December 1, 2026, for US data-centre construction figures and December 8 for power-generation figures. It gives no scheduled release date for the next METR or lab automation readings.
Would one indicator prove the scenario is happening?
No. The source says no single reading decides the scenario; its approach is to look for several indicators moving in the same direction and to account for measurement limits.
Source: ThorstenMeyerAI.com
