That is how Dan Shipper put it on Lenny Rachitsky’s podcast last week. Shipper runs Every, a media and software company that has been operating as a live experiment in AI-native work for the past two years. Every employee uses AI for almost everything. He has more direct evidence about where this is heading than almost anyone currently writing about it.
His argument is not that AI fails to automate things. It is that automation doesn’t reduce work. It raises the ceiling of what’s possible, and appetite fills the gap immediately.
The elastic ceiling
Email was supposed to reduce the time we spent on letters. It generated ten times the correspondence.
Spreadsheets were supposed to reduce the time we spent on accounting. They created a new profession of people who do nothing but build spreadsheets.
Calendar software was supposed to make scheduling faster. It produced a culture where every conversation requires a thirty-minute meeting and every meeting requires a follow-up.
The pattern is consistent: whenever we reduce the cost of something, we don’t do less of it. We discover that we wanted far more of it than we could previously afford. The demand was always there. The constraint was friction.
AI is removing friction at a scale we have never seen before. Which means the ceiling for software, analysis, strategy, creative work, and almost everything knowledge workers do is about to rise dramatically. And human appetite will fill it.
That is what automation is a lie actually means. Not that the tools don’t work. The jobs don’t disappear: they transform, expand, and move upward. The work that looked like it was about to vanish turns out to have been the floor, not the ceiling.
What actually changes
If the volume of work doesn’t fall, the distribution of effort does.
The hours that used to go into drafting, formatting, researching, and compiling now go somewhere else. The question is where. And the answer, in my experience, is that they go into the parts of the job that were always the hard parts: deciding what to do, judging whether it’s good enough, engaging with the humans who have to agree with you, and taking responsibility for the outcome.
None of those are things AI is any good at. All of them are things that were being underinvested in because the routine work kept eating the available hours.
I have been watching this play out in my own work. Workflows that used to take a team half a day now take an hour. The half day did not become free time. It became space for thinking about what we should be doing that we were not doing before, because we were too busy doing the thing that now takes an hour.
The constraint has shifted. Before, it was throughput. Now, it is judgment. And judgment is a very different problem to manage.
The forward deployed engineer
Shipper’s sharpest prediction is about a role that barely has a name yet: the forward deployed engineer.
Part engineer, part product manager, part consultant. Someone with enough technical depth to build things, enough commercial context to know what’s worth building, and enough communication ability to operate at the boundary between the two. This is the person AI companies are discovering they desperately need when they go to enterprise clients: not the sales deck and the demo, but someone who can sit inside the client’s problem and build a solution in front of them.
The role isn’t new. What’s new is how much more powerful it becomes when one person with those skills can build something in a day that would have previously required a team for a month.
I wrote about this recently in the context of what agentic AI demands from product leaders. The answer is roughly the same: the value moves toward the people who can hold both sides of the equation simultaneously: technical understanding and strategic judgment. AI is not replacing that combination. It is making it rarer and more valuable.
The job apocalypse that isn’t
Shipper says the AI job apocalypse is not happening. I agree with him, but the nuance matters.
Some jobs may be eliminated: the ones that consisted almost entirely of activities AI can now perform faster and cheaper are genuinely at risk. That is real, and it is not uniform: it falls hardest on people earlier in their careers, doing narrower work, in organisations that will replace before they retrain.
But the broad prediction — that AI will hollow out the knowledge economy, leaving most professional workers without useful work to do — relies on a theory of demand elasticity that every previous technology wave has disproved. The demand for human judgment, in contexts where the cost of mistakes is high and the value of getting it right is compounding, keeps growing. The tools change what judgment is applied to. They don’t eliminate the need for it.
What I find more concerning than the job apocalypse narrative is a subtler version of it: people who use AI to do their current job faster, without ever asking what job they should be doing instead. What AI hasn’t replaced in product management gives a sense of where the floor actually is. That is the group at real risk, not because AI replaces them, but because they spend the next three years optimising for a role that quietly becomes unnecessary.
The ceiling is rising. The question worth asking is whether you are building toward it.
The automation that is coming is not going to set you free from work. It is going to set you free from the parts of work that kept you from the harder, more interesting, more genuinely difficult parts.
What you do with that is the actual question.
Leave a Reply