A few weeks ago I stood in front of a room of entrepreneurs and tried to explain what AI is doing to our work. I called the talk "Ohhh shit," after a post by Alexander Klöpping that captured the feeling exactly: the moment you grasp how much these systems can do, and what that might mean for us. Not as users. As people.

I do this for a living. I'm one of three founders of Jump; with around twenty-five colleagues we build custom software, and increasingly we build it with AI. Last month we shipped an app that would once have taken three weeks of work in five days. I'm not a sceptic shouting from the sidelines. I'm holding the accelerator.

Which is exactly why one slide in that talk wasn't about efficiency at all. It was about a French sociologist who died in 1994 and never used a computer in anger: Jacques Ellul. I'd been re-reading The Technological Society, and I couldn't shake the sense that a book from 1954 described my industry more precisely than anything written about AI last year.

Technique, not technology

Ellul's argument starts with a definition that sounds dry and turns out to be a depth charge. He isn't writing about machines. He's writing about what he calls technique:

technique is the totality of methods rationally arrived at and having absolute efficiency (for a given stage of development) in every field of human activity.

Read that again with AI in mind. Technique isn't the model weights or the GPUs. It's the relentless pursuit, across every domain, of the most efficient method — what Ellul calls "the quest of the one best means in every field." And he watches something tip over. We stop comparing the best relative means we happen to have, and start hunting "the best means in the absolute sense, on the basis of numerical calculation."

That is, almost word for word, the pitch I hear at every AI conference. Don't ask whether this task should be automated; benchmark the options and pick the optimum. The question of whether quietly dissolves into the question of how efficiently.

"No longer in any sense the agent of choice"

Here is the passage I keep coming back to. Ellul is describing what he calls technical automatism — the way the most efficient method stops feeling like a choice and starts feeling like a fact:

Technique itself, ipso facto and without indulgence or possible discussion, selects among the means to be employed. The human being is no longer in any sense the agent of choice. [...] He is a device for recording effects and results obtained by various techniques.

I felt the truth of that this year more than ever. When a tool lets you do in five days what took three weeks, the market doesn't invite you to use it. It informs you. Your competitors have already decided. Your clients now expect the five-day price. The "choice" to adopt is the choice a falling object makes to accelerate. As Ellul puts it, the one rule technical society will not tolerate is someone "impeding this technical automatism": what can be produced must be produced.

This is the part the forecasts miss. Hinton, Gates, Musk argue about how much of knowledge work AI will swallow and how soon — twenty years until no one needs to work, says Musk. I genuinely don't know, and I've stopped pretending the fog clears further than a year out. But Ellul isn't making a prediction. He's describing a logic. And the logic doesn't need superintelligence to bite. It only needs each of us, rationally, to keep choosing the one best means.

The autonomy of the system

The characteristic Ellul fears most is what he calls autonomy. Technique, he writes, "tolerates no judgment from without and accepts no limitation." It doesn't ask permission from ethics, or politics, or faith. It sets its own terms and then asks us to adapt:

there can be no human autonomy in the face of technical autonomy. The individual must be fashioned by techniques [...] by the adaptation of man to the technical framework.

That sentence is the quiet horror of it. The danger isn't a robot uprising. It's that we slowly reshape ourselves — our schedules, our skills, our sense of what's worth doing — to fit what the system rewards. Not individuals steering the process, but the process steering individuals. On that slide in my talk I called this the loss of freedom, and I used a harder word too: a kind of idolatry. A new Babel — the attempt to build a world that is fully manageable, predictable, rational. Ellul, who was a serious Christian, would have recognised the word. So would I. I believe in Jesus, not in Elon Musk's forecasts, and that's not a throwaway line. It's the only fixed point I have from which to judge a system that claims to need no judgment.

So why am I still building?

This is where it would be easy, and dishonest, to end on a clean note of refusal. Half my audience that night was quietly wishing they could be Amish — free to choose, deliberately, how they relate to each new tool. I feel that pull too. But I went home and opened my laptop and kept building, because Ellul doesn't actually offer an exit, and pretending one exists is its own kind of lie.

What he offers instead is awareness. The single most subversive act in a system that "selects among the means" for you is to keep being the agent of choice — even, especially, when the system insists there's nothing left to choose. For me that comes down to a few stubborn habits, and they're roughly what I told those entrepreneurs:

  • Ask the question the system skips. Not "can this be automated?" but "should it, here, for these people?" Efficiency is an answer, never the question.
  • Stay critical of the output. These models predict the most probable next word; that is not the same as being right. Ask for sources. Read them.
  • Protect what efficiency erodes — the slow apprenticeship of juniors, the knowledge that lives in people and not prompts, the room to be wrong on the way to being good.
  • Lead it openly. If you run a company, you're not a passenger. Talk about the changes with empathy, write down where you will and won't use AI, and build a culture that experiments without worshipping the result.

None of that stops the wave. It isn't meant to. It's meant to keep a person standing inside it — choosing, judging, refusing when refusal is right — rather than becoming, in Ellul's cold phrase, a device for recording results.

Seventy years ago he saw the shape of the thing we're now living inside. The least we can do is not pretend it's only a tool, and not pretend we have no say. The technology is genuinely astonishing. The harder, more human question is the one it can't answer for us: how do I want to relate to it?