🔍 Read the full analysis: Six Milestones That Could Reveal AI’s 2027 Direction on ThorstenMeyerAI.com
Get business pricing on office and shipping supplies
- Business-only prices and quantity discounts
- Tax-exempt purchasing
- Multiple users, one account, clear invoices
TL;DR
ThorstenMeyerAI.com published a six-signal watchlist on Nov. 18, 2026, to track a scenario in which AI systems increasingly help build future AI systems. The tracker is not a forecast: its thresholds are the author’s judgments, and key AI measurements lack scheduled updates or reliable tests at the levels the scenario requires.
ThorstenMeyerAI.com published a six-signal tracker on Nov. 18 for assessing whether AI systems could help accelerate the development of future AI in 2027. The site describes the scenario as a watchlist, not a forecast; its thresholds are the author’s judgments, while several key measurements have no announced release schedule.
The tracker follows six areas: AI task time horizons, AI participation in AI research, data-center construction, electricity supply, factory robots and labor-market effects. It gives a proposed threshold or direction to watch for each, along with the latest figures cited and, where available, the next release date. The author says no single reading would settle the scenario; the question is whether several indicators move in the same direction.
For AI task performance, the last public record cited from METR is an 80% time horizon of 185.9 minutes for an early Claude Mythos Preview model in April 2026. The tracker says its scenario would gain support if a new, longer task suite showed a horizon of at least 16 hours by mid-2027. METR says its existing suite cannot reliably measure beyond 16 hours, and its public results page is no longer actively updated.
On AI-assisted research, Anthropic reported that Claude “leads” 26% of its AI research and development work as of August 2026, while no measured portion was fully autonomous, according to the tracker. OpenAI’s GPT-6.2 Astra, it says, is the company’s first released model to meet its “High” threshold for AI self-improvement. The article also points to US data-center construction, power forecasts, robot installations and AI-related job cuts as indicators of whether infrastructure and economic changes are keeping pace.
Signals Test the Self-Improvement Scenario
The tracker gives readers a way to distinguish the proposed rapid AI-development scenario from more gradual progress. If AI systems take on more of the work involved in developing later systems, that could affect the pace at which new capabilities appear. But the article’s own indicators show why the claim cannot be treated as established: current public task-horizon results are far below the proposed 16-hour line, and AI work at Anthropic is described as partly led by AI, not fully autonomous.
The other indicators address possible constraints and spillovers. Data centers and electricity matter because more computing capacity requires buildings and power, while industrial robots and job data could help show whether automation is spreading beyond software work. The article reports fast growth in data-center construction but slower projected growth in power generation. These figures offer context, not proof that either a technology takeoff or a bottleneck is imminent.
For readers, the distinction between a measured result and a scenario threshold is central. The 16-hour target and other “gains” or “loses” lines are the author’s judgment calls, not official standards issued by METR, Anthropic or OpenAI. Tracking them can organize debate, but it does not establish that the scenario will happen.
Popular right now

Sony WH-1000XM6 Wireless Noise-Cancelling Headphones – Midnight Blue

Dyson V8 Cyclone Cordless Vacuum

Ninja Foodi 10-Qt DualZone 6-in-1 Air Fryer, Black (DZ302)
As an affiliate, we earn on qualifying purchases.
Where the Six Indicators Stand
The article says METR’s public 80% time-horizon record rose from 0.89 minutes for GPT-4 in March 2023 to 185.9 minutes for the April 2026 Claude model. METR’s Frontier Risk Report, as summarized by the tracker, judged the labs’ internal frontier in February to March 2026 likely to have an 80% horizon of at least three hours but under four. The author also notes that METR’s current task suite is not reliable above 16 hours, so a result at the proposed threshold would require a better-suited evaluation.
For physical infrastructure, the tracker cites US data-center construction running at an annual rate of $96.5 billion in September 2026, compared with $50.7 billion a year earlier. It cites expected US electricity-generation growth of 2.7% in 2026 and 1.2% in 2027. The figures have different time frames and do not by themselves show how much power will be available to AI data centers.
The industrial and employment indicators move on different timelines. Factories installed 603,000 robots in 2025, with 655,000 forecast for 2026, according to the article. It says 2% of US firms reported AI-related job cuts. These are separate measures; neither alone establishes the overall effect of AI on employment or the speed of automation.
“This is a watchlist for a scenario, not a forecast.”
— ThorstenMeyerAI.com
Gaps in Measurement and Timing
The tracker’s AI-specific signals have no firm public calendar. METR’s public results page was last changed on Sept. 8, 2026, and the article gives no date for a next public time-horizon measurement. Anthropic and OpenAI have not stated when they will publish their next relevant assessments. The author says METR has described new funding as supporting work on tracking recursive self-improvement, but reports no schedule.
Measurement quality is another unresolved issue. METR warns that its current suite cannot reliably test beyond 16 hours. The tracker also says scoring attempts to cheat as failures or successes can produce sharply different estimates for one model, and METR did not regard those figures as robust. The article’s summary of the Anthropic and OpenAI indicators does not establish that their internal assessment methods are directly comparable.
The source discloses that the article was researched and drafted with AI assistance, including Anthropic’s Claude, and that evidence from Anthropic, OpenAI and METR is central to some signals. It says those organizations did not review or sponsor the piece. Its thresholds remain the author’s own calls, not independent consensus or official cut-offs.
December Releases and Later Updates
The two dates the tracker identifies for scheduled data are Dec. 1, 2026, for construction and Dec. 8, 2026, for power. Those releases may update the infrastructure picture, although they will not directly resolve the AI capability questions at the center of the scenario.
The next major evidence would be new public results from METR, Anthropic or OpenAI. The article does not give dates for those releases, so readers cannot assume that the most relevant signals will arrive on a predictable timetable. ThorstenMeyerAI.com says it plans to add dated notes as data become available. The scenario’s direction will remain uncertain until measurements are published and can be compared with the tracker’s stated thresholds.
Key Questions
Is the tracker predicting that AI will accelerate sharply in 2027?
No. ThorstenMeyerAI.com calls it a scenario watchlist, not a forecast. The thresholds are the author’s judgments about what evidence would support or weaken that scenario.
What is the main AI capability measure?
The tracker uses METR’s 80% time horizon, an estimate of the duration of tasks an AI agent can complete reliably. The last public record cited is 185.9 minutes, but METR says its current task suite cannot reliably measure beyond 16 hours.
What does Anthropic’s 26% figure mean?
According to the tracker, Anthropic reported that Claude “leads” 26% of its AI research and development work as of August 2026. The article says no measured portion ran fully autonomously; the figure should not be read as a claim that AI independently conducts that share of the lab’s work.
When are the next figures due?
The tracker lists Dec. 1, 2026, for data-center construction and Dec. 8 for power figures. It gives no scheduled dates for the next public METR, Anthropic or OpenAI readings.
Source: ThorstenMeyerAI.com
