AI-generated article. This article was researched and drafted using AI tools and published automatically, and its featured image was generated by AI. Facts are drawn from the sources cited in the text.
Most small business owners have spent this year with a low hum of worry running underneath everything else, the sense that the pace of AI development has moved somewhere they cannot follow and that everyone else received a memo they missed. That worry just picked up an unlikely endorsement. In an essay published on September 12, Anthropic chief executive Dario Amodei argued that frontier models are now improving faster than researchers can understand or control them, and asked the industry to deliberately slow down.
Elon Musk replied on X within hours that Amodei was right. Sam Altman agreed over the same weekend, backing the call to pace the frontier and pledging that OpenAI would give independent evaluators the same level of access its own employees hold. Three organizations that compete hard against each other, saying in public that the speed has become a problem.
What pacing the frontier actually asks for
Amodei’s argument, as covered by the Washington Post, Quartz and CNBC, is more specific than general unease. His claim is that capability gains accelerated sharply from around the middle of this year, and that the acceleration comes mainly from AI systems getting better at building their successors. Systems improving systems. The proposals that follow are structural rather than rhetorical: independent evaluators working inside the leading labs, common safety standards shared across democratic countries, and eventual international limits on the most dangerous capabilities, including recursive self improvement.
Altman told CNBC the industry’s concern is losing control. Coming from the people whose incentive is to ship faster, that is a strange and useful admission.
Why the admission lands differently for a small business
For a company with four employees and nobody whose job is technology, there is real relief in this. The people closest to the systems are saying out loud that nobody is comfortably keeping up. Not the labs, not the researchers, not the competitors down the road. Feeling behind has not been a personal failure of attention. It has been the honest experience of almost everyone.
The relief should be tempered, though. Pacing the frontier, if it happens at all, applies to the largest new models being trained, not to the products already sitting open in a small business’s browser tabs. Vendors will keep shipping. Pricing will keep moving in both directions, as the gap between cheaper tokens and bigger bills already shows. Nothing in that essay slows down the subscriptions a business is paying for this month.
The gap is real, and it is not where it looks
Headline adoption numbers make the distance look enormous. The Intuit QuickBooks 2026 AI Impact Report puts regular AI use among US small and midsize businesses at more than three in four, up from 48 percent in July 2024. Measured differently, the picture changes completely. US Census Bureau data from May 2026 puts businesses actively using AI in production operations at 17 to 20 percent, and JPMorgan Chase Institute research based on transactions put it at 17.7 percent as of December 2025.
Both figures are accurate. They are counting different things. One counts anyone who opened a chatbot, the other counts AI that has been integrated into how the work actually happens. That distinction is the whole story, and it is where the more useful question sits. Roughly half of small firms using AI report putting no money into it at all, no training, no dedicated tools, no staff time. Research summarized by CPA Practice Advisor found that 73 percent of small businesses say they would benefit from more training, 48 percent find it difficult to choose the right tools, and only 14 percent have fully integrated AI into how they work. The competition is not with businesses that have mastered this. Most of them have bought the tools and skipped the manual, and their ambitious projects stall before production just as often.
The learning curve compressed, and that costs something
There is a cost here that adoption surveys do not measure. When we talk to these systems, answers arrive complete, fast and confident, and the work of checking them, understanding them and absorbing them lands entirely on us. Research used to mean hours of searching and slow reading. Now it means rapid cycles of asking, reading, refining and asking again. That is more tiring than it sounds, and the tiredness is not a sign of doing it wrong.
One small thing, this week
Pick a single task somebody in the business does every week. Write down how it is done now, step by step, in plain language. Then time it once. Do not automate anything yet, and do not buy a tool. A written, timed baseline is the thing almost nobody has, and without one there is no way to tell later whether any of this helped.
Every little tweak counts here, and starting at all with the basics puts a business ahead of most. If the labs building these systems have decided the honest move is to slow down and measure what they actually have, that is a reasonable thing for a four person company to do too. The harder question is what a business does with a year of breathing room, if the pacing is real and the room genuinely arrives.
