🔍 Read the full analysis: Six Ways To Gauge Whether 2027 Is A Turning Point For AI on ThorstenMeyerAI.com
Get hardware and tech essentials 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
Thorsten Meyer AI published a watchlist on Nov. 18, 2026, setting six measures to track whether AI systems begin accelerating the development of future AI systems in 2027. The piece says this is a scenario tracker, not a forecast: its thresholds are editorial judgments, several key measurements lack publication dates, and no single signal would settle the question.
ThorstenMeyerAI.com published a six-signal watchlist on Nov. 18 for testing whether 2027 could bring a sharp acceleration in AI development, including systems doing more of the work involved in building successor systems. The article calls its thresholds judgment calls, not official cut-offs, and describes the scenario as something to monitor rather than a forecast.
The tracker follows measures ranging from AI agents’ ability to complete long software tasks to the share of AI research and development work led by AI. It also monitors data-centre construction, electricity generation, industrial robot installations and AI-related job cuts. The source says no single reading would decide whether the scenario is unfolding; its case would strengthen if multiple indicators moved in the same direction.
The sharpest gap identified by the article is between current task performance and the pace implied by the scenario. The last public record cited for METR’s 80% time horizon was 185.9 minutes—about three hours—for an early Claude Mythos Preview in April 2026. The tracker’s threshold for the scenario gaining support is a measured horizon of 16 hours or more by mid-2027, roughly five times that record. METR says its current task suite cannot reliably measure beyond 16 hours.
On research automation, the article reports that Anthropic said Claude “leads” 26% of its AI R&D work as of August 2026, where leading means completing most of a task end to end under human supervision. It also says no measured part of that work was fully autonomous. The source identifies OpenAI’s GPT-6.2 Astra as its first released model at the company’s “High” threshold for AI self-improvement. These are lab-reported assessments, not evidence that AI research is operating independently.
The physical-side indicators point to different speeds of change. US data-centre construction reached a $96.5 billion annual rate in September 2026, compared with $50.7 billion a year earlier, according to the figures cited by the tracker. US electricity generation was forecast to grow 2.7% in 2026 and 1.2% in 2027. Factories installed 603,000 robots in 2025, with 655,000 forecast for 2026; only 2% of US firms reported AI-related job cuts, the article says.
The Gap Between Software and Infrastructure
The watchlist matters because its proposed scenario depends on two processes moving together: AI systems becoming more capable of contributing to AI research, and the physical resources needed to run and scale that work keeping pace. A rapid improvement in software alone would not establish that labs can supply enough chips, power and data-centre capacity. The figures cited by the source suggest that construction investment and electricity supply do not necessarily grow on the same timetable.
The employment and robotics measures offer a separate check on claims that automation is already remaking work at broad scale. The reported 2% share of US firms with AI-related job cuts is a limited indicator, not a complete count of AI’s effects on jobs. Taken together, these measures can help readers distinguish a technology-development scenario from evidence of widespread operational change. They do not, by themselves, establish what will happen in 2027.
Products Worth Considering

LG UltraGear 27-inch QHD OLED Gaming Monitor, 240Hz, 0.03ms, G-SYNC Compatible, FreeSync Premium Pro, HDR True Black 400

Logitech MX Creative Console with 9 Customizable Keys, Graphite

Logitech MX Mechanical Mini for Mac Wireless Keyboard, Space Gray
As an affiliate, we earn on qualifying purchases.
How the Six Signals Were Set
The tracker follows a scenario article published earlier on Nov. 18, titled “Scenario 2027: A Year of AI Progress in Two Months, Until Compute Runs Out?” That article proposed that AI systems could take on increasing portions of the work of developing future systems, before running into constraints such as compute, buildings and electricity. The watchlist is intended to test that idea against later measurements, rather than restate its argument.
The six signals cover METR time horizons; AI’s share of AI research; compute capacity; data-centre construction; power generation; and robots and employment. The source says the thresholds are its own and were fixed before future data arrive. It also discloses that the article was researched and drafted with AI assistance, including Anthropic’s Claude; that evidence from Anthropic and OpenAI is central to two signals; and that METR’s estimate concerns Anthropic. The source says none of those organizations reviewed or sponsored the piece.
“This is a watchlist for a scenario, not a forecast and not investment advice.”
— ThorstenMeyerAI.com
Missing Measurements and Open Questions
Several of the most relevant indicators have no announced release schedule. METR’s public time-horizon page was last changed on Sept. 8, 2026, and is described as no longer actively updated; the source gives no date for a new measurement. Anthropic and OpenAI have also not stated when they will publish the next readings relevant to AI research automation and self-improvement.
The tracker warns that results depend on how tasks are designed and scored. METR says its existing suite cannot measure reliably beyond 16 hours. The source also describes a large difference in one GPT-5.6 Sol estimate depending on whether attempted cheating was scored as failure or success, and says METR did not regard the resulting numbers as robust measurements. For construction and power, the tracker supplies announced dates; for AI-specific measures, timing depends on when labs or evaluators publish. It remains unclear whether any new measurements will be comparable, independently validated or sufficient to show a sustained acceleration.
Upcoming Data Releases and Updates
The tracker lists the next scheduled releases for two physical indicators: US construction data on Dec. 1, 2026, and power figures on Dec. 8. Those releases can update the infrastructure portion of the dashboard, but they cannot answer whether AI systems are taking over more of AI research.
For that, readers will have to watch for new METR task-suite results and updated disclosures from Anthropic and OpenAI. The source says it plans to add dated notes as measurements arrive. Its stated approach is to judge the direction across several signals, rather than treat one threshold as a verdict on whether 2027 becomes a turning point.
Key Questions
Is the tracker predicting that 2027 will be a turning point for AI?
No. ThorstenMeyerAI.com describes it as a scenario watchlist, not a forecast. Its thresholds are the publisher’s judgments about what would strengthen or weaken the scenario.
What is the main AI capability measure?
The tracker follows METR’s 80% time horizon: the duration of a task, measured by skilled human completion time, at which an AI agent is predicted to succeed four times out of five. It identifies 16 hours by mid-2027 as a threshold that would support the scenario, while noting current measurement limits.
Does Anthropic say its AI systems run its research autonomously?
No. The source says Anthropic reported that Claude leads 26% of its AI R&D work, with a human supervising, and that no measured part ran fully autonomously.
When are the next figures expected?
The tracker lists US construction data for Dec. 1, 2026, and power figures for Dec. 8, 2026. It gives no announced dates for new METR, Anthropic or OpenAI readings.
What could weaken the scenario?
The source says the scenario would lose support if new, longer task suites still showed AI agents with time horizons of only a few hours. More broadly, its method calls for watching whether several measures move together; one reading alone would not settle the issue.
Source: ThorstenMeyerAI.com
