AIThis post was created with the assistance of artificial intelligence (AI).

🔍 Read the full analysis: Could 2027 Mark AI’s Takeoff? Six Signals Worth Watching on ThorstenMeyerAI.com

Buying for a business?Offer from Amazon

Get business pricing on monitors, keyboards and dev gear

  • Business-only prices and quantity discounts
  • Tax-exempt purchasing
  • Multiple users, one account, clear invoices
As an affiliate, we earn on qualifying purchases.

TL;DR

A November 18, 2026, article from ThorstenMeyerAI.com lays out six signals to monitor for a scenario in which AI helps accelerate the development of more capable AI systems in 2027. The piece is a tracker, not a forecast: several indicators lack scheduled updates, and the evidence so far does not establish that the scenario is underway.

ThorstenMeyerAI.com published a six-signal watchlist on November 18 for tracking whether AI development could accelerate sharply in 2027, as AI systems take on more of the work involved in building future AI. The article describes its thresholds as the authors’ judgment calls, not a forecast; it reports mixed current evidence and says no single indicator will settle the scenario.

The tracker follows measures of AI agents’ performance on lengthy tasks, AI’s role in laboratory research and development, and the capacity to build and power data centres. It also includes industrial robots and AI-related job cuts. The source says the scenario would involve AI systems doing more work on the development of later systems, with physical constraints such as chips, buildings and electricity potentially limiting the pace.

One key benchmark is METR’s estimate of the task length an AI agent can complete with 80% reliability, measured against the time a skilled human takes. The last public record cited is 185.9 minutes for an early Claude Mythos Preview in April 2026. The tracker’s threshold for the scenario gaining support is a measured horizon of at least 16 hours by mid-2027. METR’s public page was last updated September 8, and the source says METR’s current task suite cannot reliably measure beyond that threshold.

For AI research work, the article reports Anthropic’s August 2026 finding that Claude “leads” 26% of the company’s AI R&D work, where leading means completing most of a task end to end with human supervision. The tracker says no measured part of that work runs fully autonomously. It also identifies OpenAI’s GPT-6.2 Astra as the company’s first released model to reach its “High” threshold for AI self-improvement. These are company assessments, not independent demonstrations that AI research has become autonomous.

At a glance
analysisWhen: Published November 18, 2026; the tracke…
The developmentThorstenMeyerAI.com published a six-signal tracker on November 18, 2026, for testing its proposed scenario of sharply accelerating AI progress in 2027.

Signals That Could Test the Scenario

The tracker gives readers a way to distinguish a striking forecast from observable developments. If longer task horizons, greater AI participation in research, and expanding computing capacity all move in the direction identified by the article, that would lend support to the proposed acceleration. If they do not, or if measurements remain limited, the scenario would have less evidence behind it.

The physical indicators matter because software advances depend on infrastructure. The source reports US data-centre construction running at an annual rate of $96.5 billion in September 2026, up from $50.7 billion a year earlier. It also cites expected US power-generation growth of 2.7% in 2026 and 1.2% in 2027. Those figures point to different rates of change in construction and electricity supply, but by themselves do not establish how much compute will be available to AI developers.

For workers and businesses, the tracker cautions against treating automation as an immediate, economy-wide shift. It cites 603,000 factory robot installations in 2025, with a forecast of 655,000 in 2026, and says only 2% of US firms reported AI-related job cuts. Those observations offer context, not proof that AI has or has not changed employment broadly.

Products Worth Considering

Logitech MX Mechanical Wireless Illuminated Keyboard, Tactile Quiet, Graphite

Logitech MX Mechanical Wireless Illuminated Keyboard, Tactile Quiet, Graphite

As an affiliate, we earn on qualifying purchases.

How the Six-Signal Watchlist Works

The article follows an earlier scenario analysis, “Scenario 2027: A Year of AI Progress in Two Months, Until Compute Runs Out?” The November 18 piece does not repeat the argument for that scenario. Instead, it specifies what to measure, identifies data publishers and available timing, records the status as of publication, and sets lines at which the scenario would gain or lose support.

The six indicators cover AI task horizons; AI’s share of AI research and formal self-improvement assessments; data-centre construction; power generation; industrial robot installations; and AI-related employment cuts. The source says its thresholds are not official cut-offs. For compute and labour, it does not set a numerical threshold and instead asks readers to follow the direction of the series. The article says the scenario should be judged by whether multiple signals move together.

Only two data releases have dates in the source: US construction figures on December 1, 2026, and power-generation figures on December 8. The AI-specific readings depend on when METR and the AI companies publish new information, and the source gives no schedule for them.

““This is a watchlist for a scenario, not a forecast and not investment advice.””

— Thorsten Meyer, ThorstenMeyerAI.com

Key Data Gaps and Limits

The tracker’s central uncertainty is whether the observed pace of AI progress can approach the scenario’s implied acceleration. The source says METR estimates Anthropic is improving models on well-specified tasks at about twice its earlier rate, while METR “would not be surprised” by a rate of 1.3 times or three times. The tracker compares those estimates with a roughly sixfold pace implied by a year of progress arriving in two months; the available figures do not establish that this faster rate has been reached.

METR’s last public time-horizon record is several months old, its public page is no longer actively updated, and the current task suite has measurement limits. The source also describes a large difference in one METR result depending on how suspected cheating is scored, and says neither resulting figure is robust. Comparisons require caution, particularly until longer tasks can be evaluated reliably.

Other gaps include the absence of publication schedules for lab self-assessments and METR’s future readings. The source material supplied for this article also does not fully detail the current status or precise thresholds for every signal, including some of the robot and employment measures. The construction and power figures are infrastructure indicators, not direct measures of AI capability or available AI compute.

Upcoming Releases and Lab Updates

The next dated releases identified by the tracker are US data-centre construction figures on December 1, 2026, followed by power-generation figures on December 8. Those updates can show whether recent infrastructure trends continue, though they cannot alone confirm the proposed AI-development scenario.

For the capability and research indicators, readers will have to wait for METR and the labs to publish new measurements or assessments. ThorstenMeyerAI.com says it plans to add dated notes as data arrive. The scenario’s evidence will depend on the quality and comparability of those measurements, as well as whether several independent signals point in the same direction.

Key Questions

Is the article predicting that AI will take off in 2027?

No. The source calls the piece a watchlist for a scenario, not a forecast. Its thresholds are the authors’ judgments about what would strengthen or weaken that scenario.

What is the main capability signal?

The tracker focuses on METR’s 80% task time horizon: how long a software task can take a skilled human while an AI agent still completes it successfully four times out of five. It sets 16 hours by mid-2027 as a line that would support the scenario, while noting current measurement limits.

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

No. In the source, “leads” means Claude completes most of a task end to end while a person supervises. The article says no measured part runs fully autonomously.

When will the next data arrive?

The source identifies US data-centre construction figures due December 1, 2026, and power-generation figures due December 8. It gives no scheduled dates for the next METR or lab assessments.

What would count as evidence for the scenario?

The article does not treat any single reading as decisive. It says the scenario would gain support if multiple signals, including longer reliable task horizons and rising AI participation in AI research, move in the expected direction; weak or flat readings would weigh against it.

Source: ThorstenMeyerAI.com

You May Also Like

My Multi-Mac Setup With Claude, Codex, And T3 Code

Thorsten Meyer says he moved eight Macs and a Linux server into one T3 Code workspace for managing Claude Code and Codex sessions.

40% Performance Boost? Claude AI Creates An Oblivion Remastered Mod

A TweakTown headline says Claude helped create an Oblivion Remastered mod with a 40% performance boost, but benchmark details are unavailable.

Claude Fable 5.2 Costs Per Task: What To Expect At Each Effort Level

Artificial Analysis reports Fable 5.2 scores from 54 to 65 across five effort levels, while average task costs rise from $3.10 to $26.54.

Pay For Hugging Face Hub Through Your Google Cloud Account

Hugging Face organizations can link a Google Cloud billing account to pay for eligible subscriptions and metered usage through Marketplace.