The Bannister Effect
AI and the shifting edge of impossible
Here’s a weird thought I can’t shake: an amateur British runner from 1954 explains the AI moment better than most of the people writing about it.
His name was Roger Bannister, and for a very long time a lot of serious people had built a lot of serious arguments for why what he was about to do was impossible. Running a mile in under four minutes. The human body had limits. There was a wall. Then on May 6, 1954, Bannister went out and ran one in 3 minutes and 59.4 seconds. Forty-six days later John Landy broke the record too, and within three years sixteen other runners had done it. The wall was never physical. It was psychological, and once everyone could see that the thing was possible, the impossible quietly became routine.
I’ve been thinking about this a lot lately, because I’m pretty sure we’re all living inside a version of that story right now, except it’s playing out across a dozen different domains at once and at a pace that’s honestly hard to keep up with.
I was a history major, so I come at this with a particular bias, but the Bannister pattern is not unusual. It’s basically the shape of how progress actually happens. Someone breaks the sound barrier, and suddenly everyone is breaking it. Someone sequences the human genome, and a decade later your cousin can spit in a tube for $99. The gap between “impossible” and “commodity” is almost always shorter than anyone predicts, and the thing that closes it is rarely a new law of physics. It’s usually one person making the invisible wall visible by walking through it.
What’s strange about this particular moment is that we’re watching the pattern play out on multiple fronts at the same time. Think about the last eighteen months. AI is writing code that senior engineers are actually shipping to production and feeling good about, which was basically a punchline not that long ago. People are letting autonomous agents run real pieces of their work and their lives, making decisions without a human signing off on every step, which sounded like sci-fi a year ago. You can run a full, capable AI model locally on the phone in your pocket with no cloud connection at all, which people were writing papers about being decades away. Every one of those was on somebody’s list of impossible things, and none of them are anymore.
The weirder part is that most of us haven’t had time to update the list. We’re all still walking around with a mental inventory of impossibles that was written before any of this existed. The list still says AI can’t do real reasoning, non-technical people can’t build real software, automation is a developer thing, complex integrations need an engineer, and so on. These were all reasonable positions at some point, and the muscle memory of believing them is strong. But the walls keep coming down, model release by model release and skill by skill, and the possible is expanding faster than most of us are editing the list in our heads. A lot of the conversations I have start from assumptions about what AI can’t do that stopped being true about six months ago. And the breakthroughs keep compounding on each other. One thing working makes the next thing easier to imagine, which makes it easier to try, which makes it more likely to happen. We are not in a single Bannister moment. We are in a Bannister loop, and it is running fast.
The thing I love most about working in this field right now is that it’s not just a story about researchers and labs. It’s a story about everyday builders, and at Zapier I get a front row seat to it. I work alongside people who are building genuinely amazing things, and I talk to customers who sit down to solve some annoying part of their job and end up shipping automations and AI-powered workflows that, a year or two ago, would have needed an engineer, a project plan, and a budget. A lot of these folks would have told you, unprompted, that they weren’t technical. They were wrong about themselves, and watching them figure that out in real time is one of the best parts of the job.
The part of the Bannister story nobody focuses on is what happened after Landy. The next wave of runners who broke four minutes weren’t some new breed of superathlete. They were regular runners who now knew, for a fact, that it could be done, and trained accordingly. The shift was in the belief. Every time someone who “isn’t a builder” ships something useful with AI, they’re doing the same thing Bannister did on a smaller scale. They run through the tape, and the crowd behind them quietly updates what they think is possible for themselves.
So what am I watching for next? Honestly, I don’t know exactly which wall is about to come down, and I’ve learned to hold any prediction I make loosely because I’ve already seen enough things on my own list of impossibles turn out to be wrong. What I do know is that the pattern is real, and that the right people, the right technology, and the right mindset are colliding right now across more domains than any single person can track. Most of what’s on your list of impossible is just sitting there waiting for its May 6, 1954.
I’m trying to be one of the people who documents that as it happens, and who helps make the breakthroughs at the edge accessible to someone who is one updated belief away from doing something remarkable. That’s a good job to have right now.
The people who win this era aren’t the ones with the best tools. They’re the ones who updated their list of impossibles first. What’s on yours? The honest, slightly embarrassing “I didn’t think I could do that” version.
Add your own entry to the list of impossibles
The story in this essay also opens my Gainsight Pulse workshop.