iTechGuides is reader-supported. When you buy through links on our site, we may earn an affiliate commission. As an Amazon Associate I earn from qualifying purchases. Learn more
In a September 15, 2026 essay, Dirk Mattig argues that people may gain more by putting today’s AI tools to work than by waiting for the next major model leap. His point is about how users and organizations spend their attention—not proof that AI capability progress has slowed. The public debate also includes a different rationale: Anthropic CEO Dario Amodei has proposed pacing frontier development to give safety work and independent evaluation time to keep up.
Mattig’s case: use what is already available
Mattig’s argument is practical and personal. He says the AI tools people can already access could keep them occupied for years as they learn where those tools help. In his view, a constant focus on the next breakthrough can distract from applying current capabilities to useful work. He writes, “An industry constantly being distracted and interrupted by the next big leap forward does not necessarily deliver productivity gains.”
That is a warning about attention and adoption, not a measured finding that current AI tools have reached a ceiling or that industry progress has slowed. Mattig also says, “I am not convinced that safety concerns are the only reason behind this AI slowdown debate.” He does not establish what other motives may be involved, and he acknowledges outsiders cannot determine how much legal, financial, or commercial considerations matter.
What “pacing” means in the safety argument
Amodei’s September 2026 essay, “We Must Pace the Frontier”, advances a policy proposal rather than Mattig’s adoption-focused argument. Amodei says companies should deliberately pace capability development so safety work and independent evaluations can keep up. His proposal includes embedded third-party evaluators and coordination at democratic and global levels.
#1 Best Overall
He explicitly distinguishes pacing from stopping progress: “To be clear, pacing does not mean halting model training or technical progress, but ensuring companies take adequate time to align and safeguard their models, and for third party evaluators to confirm this.” The distinction matters: a call to build in time for evaluation is not itself a claim that frontier development has already slowed.
What business returns do—and do not—tell us
AI adoption and frontier capability are different questions. PwC’s 29th Global CEO Survey, published January 19, 2026, collected responses from 4,454 CEOs in 95 countries and territories between September 30 and November 10, 2025. Its results point to uneven reported financial returns from AI:
| Reported outcome | Share of surveyed CEOs |
|---|---|
| Neither higher revenue nor lower costs from AI | 56% |
| Both higher revenue and lower costs from AI | 12% |
These are CEOs’ reports about their companies, not a direct measurement of productivity across all organizations or of whether AI models are improving more slowly. PwC summarized the result this way: “Most CEOs say their companies aren’t yet seeing a financial return from investments in AI.” The survey helps explain why learning to implement existing tools is a live business concern, but it cannot establish that the tools themselves have stopped advancing.
Has AI progress actually slowed?
The sources discussed here do not settle that empirical question. A lively slowdown debate, a proposal to pace development, and uneven business returns are not evidence on their own of a technical slowdown. Establishing one would require suitable measurements over time; the cited material does not provide a time series or an independent measurement of frontier-model progress.
There is also a legal story, but it should not be mistaken for proof of coordination. AP reporting published by OPB on September 20, 2026, says four paid AI subscribers filed a proposed class action in the Northern District of California on September 18. The complaint alleges that Anthropic, OpenAI, SpaceXAI, and Google coordinated to slow development. Those are plaintiffs’ allegations; the report does not describe a court finding that an agreement existed.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.What a user can take from the debate
Mattig’s closing thought is a useful prompt for individual users: “I don’t know about you, but I could use a bit of a breather.” Rather than treating each new model announcement as the only meaningful form of progress, users and teams can assess whether the tools they already have improve a real workflow. That is a practical choice, not a guarantee of productivity gains. Amodei’s proposal addresses a separate question—how to give safety work and outside evaluation time alongside continued technical progress.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
Free tools Windows power users keep installed
One-click scans. No signup required.
