Something strange happened in September. The people who spent the last few years telling us to go faster started telling us to slow down. 

Anthropic's CEO, Dario Amodei, published a long essay called "We Must Pace the Frontier", where he argued that the AI industry should deliberately ease off the speed at which it makes models more powerful, to give safety and alignment work a chance to catch up. Within hours, Sam Altman of OpenAI posted that he agreed. Elon Musk replied with three words, "Dario is right." Two or three companies that compete for the same customers and the same headlines suddenly nodding along to the same idea. That doesn't happen often.

And it left me with a question I keep turning over. Not "are the robots coming for us," which I find exhausting, but a smaller one. What if it slows down? Can it, really? What would that even mean for the world, the people who aren't building the frontier, who are down here trying to use this stuff to run a business? Because, there are many of them…

Two stories about why

There are two ways to read this moment, and the internet is busy fighting about which one is true.

The first story is the honest-fear one. The models got good enough to start helping build the next models, the pace jumped, and some of the people closest to it got scared. Amodei wrote that pacing doesn't mean pausing, it means slowing the rate of change by a year or two so the safety work isn't left behind. A researcher at his own company had quit weeks earlier, saying out loud that the labs were gambling with people's lives. If you take that camp at their word, this is a group of people seeing something the rest of us can't, and pumping the brakes before it's too late.

The second story is colder. A slowdown, conveniently, is easiest for whoever's already ahead. Raise the cost of staying at the frontier and you make life harder for the smaller labs and the open-source crowd chasing you, not for yourself. One venture capitalist put it bluntly, saying the essay was really a case to concentrate power with Anthropic. Others pointed out it lands right before a wave of enormous valuations and public offerings, and a careful, grown-up "we're slowing for safety" story protects those numbers a lot better than admitting the growth might be cooling on its own.

Here's where I've landed, and it took me a minute. Both can be true at the same time. Someone can have a real fear and a convenient incentive in the same breath, the way most of us do about most things. I don't think we get to know which one is driving, and I've stopped waiting for a clean answer. Because what matters for our work sits a level below all of that.

The AI pace was already the problem

Forget the labs for a second. Down here, where normal companies live, the speed has been quietly breaking things for a while.

Six in ten US companies say AI is being adopted faster than they can govern it. Almost half of the AI leaders in one survey admitted they'd skipped their own governance process to push something out under pressure. Read those two numbers together and you get a picture of a lot of businesses sprinting with their shoelaces tied, shipping AI into their work faster than they can check whether it's safe, correct, or even useful.

That's not a frontier problem. That's a Tuesday. It's the marketing team using a tool nobody signed off on, the finance team trusting an output nobody verified, the whole company learning a new way of working every few weeks because the thing updated again and now the old way is gone. We feel it at Solveo too. A task that used to breathe across a week gets squeezed into a day because the expectation quietly reset, and nobody decided that on purpose, it simply happened while we were keeping up.

So when I ask what if it slows down, I'm not picturing a sad, frozen industry. I'm picturing something closer to relief.

What a slower year with AI would buy us

We wrote a while back about using AI when you need it rather than because everyone is, and the whole point was that speed is not the same as progress. A slowdown, whatever its motives, would hand the rest of us the one thing the race has been stealing: time. Time to catch up, time to learn the tools properly instead of surfing the next one before we understood the last, time to put the boring guardrails in place that nobody builds when they're sprinting.

Nobody really agrees on what a slowdown would even look like, and if you want proof, go read the Reddit thread that got me thinking about all this in the first place. It's a proper brawl. One person is convinced the whole slowdown talk is a smokescreen, cover for the fact that the hype ran way ahead of what these companies can deliver, and a way to cool an overheated market before it crashes on its own. Someone else fires back that the people who understand this best have been asking for a slowdown for years, so maybe take them seriously for a minute. Others argue the models only did the scary things because humans told them to, so where's the real danger. Everyone sounds completely sure of themselves, which is usually the clearest sign that nobody actually knows yet.

That uncertainty is why a slower stretch would help. When nobody agrees on what's happening or where it's going, the worst thing you can do is move faster. You end up making big, hard-to-undo decisions in the dark, betting the business on a tool you adopted last week and barely understand. Slowing down doesn't mean stopping. It means giving yourself enough room to look before you leap, which sounds obvious until you notice how rarely any of us have done it these past couple of years.

I believe that more than I expected to. The companies that will do well with AI over the next decade aren't the ones who adopted it the fastest. They're the ones who understood it well enough to use it with some judgment, who kept a human hand on the parts that matter, who built the muscle instead of renting the output. None of that happens at a sprint. It happens in the slower stretches, the ones we keep skipping because someone made us feel like falling a week behind meant falling out of the race entirely.

So, what if AI slows down?

Maybe nothing slows down. None of the labs has announced that they're shipping fewer models, and a post on X is a much lower bar than an actual coordinated pause. The whole thing might turn out to be a lot of talk. But the fact that the loudest voices in the room all reached for the same word in the same week tells me something shifted, even if the machines keep moving at the same speed.

For years the scary story was that AI was moving too fast for us to keep up. We treated the speed as the threat. But the real cost was never only the models getting smarter. It was us, never catching our breath, never stopping long enough to ask whether we needed the thing we were reaching for.

If it slows down, that's not the disaster. That's the gift. A little room to use this well, on purpose, like people who want to still recognise how they work in a few years.

I'd take it.

Want to keep going? We've written more about AI, how we use it and what we make of it.

‍