Too Many Prophets, Not Enough Practitioners
AI discourse has too many prophets of inevitability and doom. Risk deserves seriousness, but seriousness should create agency for people doing the work.
The AI conversation has too many prophets and not enough practitioners, and the prophecy is not only on one side.
There are prophets of inevitability: the corporate, platform, and frontier-lab voices telling us this is simply where history goes now. The models will get better, the products will arrive, and the rest of us should adapt to a future mediated by whatever they’re building.
There are also prophets of doom: the catastrophe voices telling us that this is not a technology wave but an extinction machine. Stop everything. Shut it down. Anything less is complicity in the end of the species.
The tone is different, but the posture is familiar. They talk in civilizational register, claim adult seriousness, and ask for deference to their vision of the future. Meanwhile, the rest of us still have to decide what to do on Monday morning.
I don’t mean that AI risk is fake; quite the opposite. I work with these systems every day, and I think we’re dealing with a new high-level intelligence entering human society. That’s not a normal software platform cycle; it deserves seriousness.
This is where the term “AGI” mostly gets in the way.
It’s become a marketing claim, a philosophical trap, a funding signal, and an argument grenade. Everyone can define it just precisely enough to win the conversation they’re already having. Every time the term enters the conversation, the goalposts for what it means have moved.
And yet, as worthless as the term has become, I increasingly think the thing people were trying to point at is already behind us under almost any definition that matters socially. There’s a new intelligence in the room. Not a person, not a god, not magic, not an alien mind from a movie, but a real, generalizing, language-using intelligence that can write code, explain ideas, hallucinate, surprise us, and participate in work.
That should concern us, but concern is something to hold, not something that holds you.
The trick is not letting concern turn into a performance of concern. Seriousness is not the same thing as theatrical certainty.
There’s a style of AI discourse that treats certainty as the only adult position. Sometimes the certainty is “this will transform everything and you have to get on board.” Sometimes it is “this will kill everyone and if you don’t help stop it you obviously also want to kill everyone.”
Either way, ordinary engineering judgment starts to sound naive. Uncertainty becomes moral failure. Disagreement becomes proof that you haven’t understood whichever future is being sold. Ambiguity gets converted into status, and builders are left with a fake choice: submit, or be reckless.
I reject that framing - not because I’m cavalier about AI, but because I’m not.
I use them in my own work to write, code, review, debug, teach, and think. They’ve multiplied my output. They’ve also hallucinated, flattened prose, missed obvious context, and invented confidence out of thin air.
The work keeps dragging me back to boring questions: what did the system see, what did it touch, what did it cost, what can be reversed, and who is accountable when the answer is plausible and wrong? That’s not as intoxicating as prophecy, but it’s where the work is.
I’m increasingly suspicious of AI discourse that gives people nowhere to stand. If the only serious conclusion is “trust the labs, they have the future handled,” the conversation has become sales. If the only serious conclusion is despair, paralysis, or deference to prophets of abstract doom, the conversation has become liturgy. Neither helps the people who have to build with these systems and then live with the consequences.
Risk work should create agency. It should leave practitioners more capable of acting, not less.
Those questions are less grand than “are we all doomed?” They’re also more useful on a Tuesday.
We’re not going to get through this well by refusing to look at risk or by turning risk into prophecy. And we definitely won’t get through it by mistaking the business model of the frontier labs for wisdom.
The place I want more of the conversation to happen is closer to the work, where abstractions have to pay rent. Not in the sales pitch, not in the liturgy, but in the boring middle where someone has to make the system inspectable, reversible, useful, and honest about what it doesn’t know.