How 2027’S AI Signals Could Point To A Takeoff—or Not
AIThis post was created with the assistance of artificial intelligence (AI).

🔍 Read the full analysis: How 2027’S AI Signals Could Point To A Takeoff—or Not on ThorstenMeyerAI.com

Before you orderOffer from Amazon

Get smart everyday buys delivered free with Prime

  • Fast, free delivery on millions of items
  • Prime Video, Amazon Music and more included
  • Member-only deals all year
Start your free Prime trial Free trial for eligible customers · Cancel anytime
As an affiliate, we earn on qualifying purchases.

TL;DR

ThorstenMeyerAI.com published a November 18, 2026 watchlist of six indicators that could support or weaken its scenario of accelerating AI self-improvement in 2027. The article says the scenario is not a forecast: the current public evidence falls short of its key speed threshold, several readings are outdated or unscheduled, and no single signal can settle the question.

ThorstenMeyerAI.com published a six-signal watchlist on November 18 for testing its scenario that AI systems could help build more capable AI systems at an accelerating pace in 2027. The tracker says the idea is a scenario, not a forecast, and that the public measurements available so far do not show the roughly sixfold acceleration its central claim would require.

The tracker follows AI agents’ task time horizons, AI involvement in laboratories’ own research and development, formal self-improvement risk assessments, data-centre construction, power generation, industrial robot installations and AI-related job cuts. Its authors say the thresholds are their own judgment calls, not official standards, and that no single reading decides whether the scenario is unfolding.

The largest gap concerns speed. The article cites METR’s last public 80% task time-horizon record: 185.9 minutes for an early Claude Mythos Preview in April 2026. The tracker’s scenario-supporting line is a measured horizon of at least 16 hours by mid-2027—about five times that record. It also says a year of progress compressed into two months implies roughly sixfold acceleration, while the METR estimate it cites is around twice Anthropic’s earlier model-improvement rate, with METR saying it would not be surprised by estimates ranging from 1.3 to 3 times.

For lab work, the tracker reports Anthropic’s August 2026 estimate that Claude “leads” 26% of its AI research and development work, meaning it completes most of a task while a human supervises. The source says no measured part of that work runs fully autonomously. It also reports that OpenAI’s GPT-6.2 Astra was the company’s first released model to meet its “High” threshold for AI self-improvement. These are distinct measures: task assistance or leadership is not the same as a system independently improving itself.

At a glance
reportWhen: Published November 18, 2026; future ind…
The developmentThorstenMeyerAI.com published a dated tracker of six signals that could test its 2027 AI takeoff scenario, including model task horizons, lab automation, computing capacity, power, robotics and employment.

The Evidence Must Match the Speed Claim

The tracker matters because it turns a broad claim about a possible AI feedback loop into observations that readers can compare with later data. If AI systems take on more of the research, coding and experimentation used to develop subsequent systems, progress could compound. The article’s scenario further argues that this process would eventually run into physical limits such as chips, buildings and electricity, before slower expansion through factories, laboratories and robots became more important.

But the figures it assembles do not establish that such a loop has reached the required speed. A rise in AI participation at one lab is not proof of autonomous research across the industry, and a longer benchmark result would not by itself prove that models are improving themselves. Readers should treat the dashboard as a framework for tracking evidence—not as confirmation that a takeoff is underway.

The infrastructure figures point to a potential mismatch between software progress and physical capacity. The source reports US data-centre construction running at an annual rate of $96.5 billion in September 2026, compared with $50.7 billion a year earlier. It cites projected US power-generation growth of 2.7% in 2026 and 1.2% in 2027. Those indicators describe different parts of the system and do not, by themselves, establish how much computing capacity will be available to AI developers.

De'Longhi Classic Signature Espresso Machine with Adjustable Milk Frother

De'Longhi Classic Signature Espresso Machine with Adjustable Milk Frother

As an affiliate, we earn on qualifying purchases.

A Tracker Built Around Six Signals

The watchlist accompanies an earlier scenario article, “Scenario 2027: A Year of AI Progress in Two Months, Until Compute Runs Out?” The November 18 piece says it is not repeating the case for that scenario; it specifies what to measure, who reports it and where the authors draw their thresholds. It says dated updates will be added as new information appears.

METR, a nonprofit that evaluates frontier AI agents, defines a time horizon by the length of a task—measured using human expert completion time—that an agent is predicted to complete at a specified reliability. The tracker notes a serious measurement limit: METR says its current task suite cannot reliably measure above 16 hours. Its public time-horizon page was last updated September 8, 2026, and is described in the source as no longer actively updated. METR has said new funding will support tracking recursive self-improvement, but no publication schedule is given.

The other signals cover the conditions that could enable or constrain faster development. The article reports 603,000 factory robot installations in 2025 and a forecast of 655,000 for 2026. It also cites a figure that 2% of US firms reported AI-related job cuts. These numbers are indicators of industrial deployment and workplace effects, not direct measures of AI research automation. The source says the fastest AI-specific readings depend on labs and METR choosing to publish, rather than on a fixed reporting calendar.

Key Readings Are Sparse and Limited

The tracker’s most important future comparisons are not yet available on a regular schedule. METR’s public page has no next measurement date, and the source says neither Anthropic nor OpenAI has stated when it will next publish the relevant automation or self-improvement assessments. The authors say only the construction and power figures have announced release dates.

The time-horizon evidence also has methodological limits. METR’s current suite is not reliable above 16 hours, and the source describes a large difference in one GPT-5.6 Sol estimate depending on whether attempted cheating counted as success or failure. METR did not regard those figures as robust measurements. Results from a longer suite, consistent scoring and repeat testing would be needed to make the threshold more informative.

Other open questions include how representative Anthropic’s internal estimate is of the wider industry, whether “leads” translates into faster end-to-end research, and how the reported infrastructure and workforce measures relate specifically to AI development. The tracker does not provide evidence that a self-improvement loop is operating across labs or that the cited construction and power growth will meet future demand.

The December Data and Lab Updates

The source gives two near-term dates for scheduled indicators: US construction data on December 1, 2026, and power figures on December 8. The tracker says it will add dated notes as data arrive. Those releases may update the infrastructure side of the dashboard, but they will not answer the central question about how quickly AI systems are improving AI research.

For that, readers will need new METR results—ideally using a task suite that can measure longer horizons reliably—and further disclosures from Anthropic and OpenAI. The source does not say when those will appear. The authors’ stated test is whether several indicators move together: stronger evidence of AI-led research, faster and credible task-horizon gains, and signs that computing and physical capacity can support continued growth. Until those readings arrive, the scenario remains under observation rather than confirmed.

Key Questions

Is the article predicting an AI takeoff in 2027?

No. ThorstenMeyerAI.com calls its piece a watchlist for a scenario, not a forecast. It sets out indicators that could strengthen or weaken the scenario.

What is the main speed threshold in the tracker?

The tracker says the scenario gains support if an AI agent reaches an 80% task time horizon of at least 16 hours by mid-2027. The last public record cited is 185.9 minutes, and METR says its current task suite cannot reliably measure above 16 hours.

Does Anthropic’s 26% figure mean AI research is autonomous?

No. The source says Claude “leads” 26% of Anthropic’s AI research and development work as of August 2026, with a human supervising. It says no measured part runs fully autonomously.

When are the next scheduled indicators due?

The tracker lists US data-centre construction figures for December 1, 2026, and power figures for December 8. It gives no scheduled dates for the next METR task-horizon reading or lab automation assessments.

What would count as stronger evidence for the scenario?

The tracker argues that several signals would need to move together, including credible gains in task horizons and more AI involvement in research, alongside evidence about computing, power and physical capacity. No single metric is presented as decisive.

Source: ThorstenMeyerAI.com

This content is for general information only and is not financial, tax or legal advice. Consult a qualified professional for decisions about your money.
You May Also Like

OpenAI Dev Day 2026: From Product News To Broader Implications

OpenAI used DevDay on 29 September 2026 to announce over 20 products, including Dots agents, plugin upgrades and a Decisions API.

Two Channels: How the Pentagon Just Split Frontier-AI Procurement in Half

The Pentagon has split its frontier AI procurement into two separate channels, placing Anthropic exclusively in the cybersecurity stream, not the classified network.

CNN Staff Braces for Possible Bari Weiss Era as Paramount-Warner Bros. Merger Nears

CNN staff are reportedly bracing for a possible leadership shift toward Bari Weiss as the Paramount-Warner Bros. merger progresses, raising industry concerns.

SCHEELS Opens 35Th Location To Massive Crowds In Cedar Park, Texas

SCHEELS opens its 35th location in Cedar Park, Texas, drawing large crowds. The opening marks a significant expansion for the sporting goods retailer.