Another major AI vibe shift is happening.
The tech is moving so fast that our collective reactions are emotionally exaggerated.
In mid 2022 most people didn’t think anything could pass the Turing test. We got chatbots in 2023. We got agents in 2024, but nobody trusted AI for coding.
We got Claude Code in 2025, and in one calendar year we went from “I’d never let AI code for me,” to “I let it write the code but I review it,” to “I don’t even write code anymore.”
In late 2025 and early 2026 most boards and CEOs started demanding everyone use AI. For what, they weren’t quite sure. The vibe was “we need to immediately replace workers with AI.” People thought everything would be AI inside companies within 18-36 months.
And now in mid 2026 the pendulum is coming back. “Oh crap, that’s hard and expensive. We better slow down and rethink this.” People start thinking this was a giant miss, and that human workers will never go away.
The swing from “it can’t code” to “I don’t code anymore” happened in less than a year. And now this one has gone from “AI will replace all human workers” to “AI can’t replace human workers” in less than six months.
A cancer diagnosis
Pardon the cringe analogy, but I’m reminded of when someone is dying of cancer. In this case I think the thing dying would be the old way of work being done.
When someone gets a major or terminal cancer diagnosis there is tons of uncertainty. You keep testing and testing, and you often end up with two groups of people.
We’ve already been told it’s terminal, but whenever there are positive responses to a treatment, or positive test results, one group concludes: “It looks like we beat this thing. You’ll probably live another couple of decades.”
And there’s another group who, upon hearing any negative result or analysis, says they could pass any day, or any week. They’ll be gone in less than a few months.
I’ve seen this structure play out multiple times.
- It’ll happen in less than 2 months.
- It’ll never happen.
And sometimes people in one group switch to the other group.
But what I’ve seen basically every time is that it ends up being something in the middle. It doesn’t happen instantly like the fast group said. And then, right when it starts to look like things are going to be fine, it happens.
Somewhere in the middle.
I hate this analogy for obvious reasons, but I feel like we’re collectively stuck in this mode when it comes to the inevitability of change from AI. The combination of “really bad” with “don’t know when” causes a special kind of anxiety and reaction in us humans.
What’s actually at risk
This model doesn’t necessarily tell us what’s happening in AI, because it depends on what you use as the substitute.
My intuition is that the thing at risk in our current situation is “most work being done by humans.” That is the thing that now has limited time.
You can find all over Twitter the people who think 20% unemployment or whatever will happen in 1-2 years. And an equal number of people who think this is a psychosis that will soon end. Or that it’ll take 10 or 20 years instead.
This reminds me a little of regression to the mean. It’s something like: find the median number for all the guesses.
But that method wouldn’t have been correct in early 2022, so who knows.
I just think this mental model might be useful in thinking about what could happen, and how long it might take.
