AIpocalypse, Just-in-Time
The world is ending again. This time it is AI that will do us in. Superintelligent machines by next year, humanity eclipsed by the end of this one, the singularity arriving after decades of being five years away. We have been using these tools daily for more than two years now, and the discourse around them has completely detached from the experience of actually using them. That gap is worth understanding, because it has real consequences for how businesses decide to invest.
The eschatological assembly line
The world has been ready to end many times in recent memory. The various predicted apocalypses, the rapture that keeps getting rescheduled, the Y2K meltdown, the 2012 calendar, the singularity. The structure is identical each time. A transformation is always imminent, never arrives, and the faithful are expected to maintain urgency indefinitely. When the predicted end does not come, the community does not question the prediction. It revises the date and intensifies devotion. The date coming and going is never evidence the idea was wrong. It just means someone miscalculated. The AIpocalypse is the latest product off this assembly line.
Just-in-time manufacturing
The phrase just-in-time comes from logistics. Factories stopped keeping warehouses full of parts and instead ordered components to arrive exactly when needed, driving inventory costs toward zero. The attention economy learned the same trick. The news cycle cannot tolerate a vacuum, so there is always a crisis in transit, arriving just as the previous one expires. The AIpocalypse is not a long-term forecast grounded in evidence. It is a perishable good manufactured this morning to be consumed by lunch. By dinner there will be a fresh shipment. Valuations need it. Career positioning depends on it.
Set that against reality. One of us built a personal website in a language and framework he had never shipped anything in, going from zero to one with an AI coding assistant. That was most of a year ago. Since then, the experience has plateaued. People are doing impressive things, but it is all engineering now: agents orchestrating agents, normal software engineering principles applied to AI tooling. The hype from AI labs follows an exponential curve. Our commit history follows a linear one. Both of those things can be true at the same time, and only one of them shows up in the work.
The content mill was already running
AI is not special in generating a parasitic content ecosystem. The ten tips for prompting is the same pattern as ten command-line tricks or ten spreadsheet shortcuts. The content engine was already running. AI is just the current fuel. The professional networks are dominated by it. Five patterns you should use. Spot the bad pattern. Competence farming, where the post deliberately mixes good ideas with bad ones, the experts flood the comments to correct it, and the engagement feeds the machine regardless. There is also the steady stream of non-technical people giving technical advice. What is being imitated in all of this is not proficiency. It is the performance of proximity to the hype. You perform urgency about the imminent transformation, and you audition for a role in someone else's manufactured drama.
What the tool actually does
Here is the line that matters for anyone deciding how to invest. AI can generate. It cannot evaluate. We once did a deep dive into a legacy application for a client, using multiple agents to validate each other and spot-checking for hallucinations. When the agents were wrong, they were usually right about the general direction and wrong about the specifics. The pattern existed and was bad, but the count was off. A human had to verify that the flagged patterns were actually problems. The meaning required human judgment.
This is why expertise still matters and why the panic about knowledge workers becoming obsolete is overblown. You need to know enough to verify what the AI generates, to recognize when it is confidently doubling down on a wrong answer, when the direction is right but the count is off. The AI does not know it is wrong. It cannot. Human beings are the ultimate creators of meaning. AI produces a pattern of tokens. We are the ones who recognize that a particular pattern has meaning, that a flagged issue is a real problem, that an output is worth acting on. Strip the human out and you have not automated the work. You have automated the production of plausible output that nobody is qualified to trust.
Refuse the shipment
Once you recognize the pattern, you can stop performing. The apocalypse keeps getting manufactured because we keep buying it. The attention economy presents whatever riles you, the factory floor never stops, and there is always another micro-crisis in transit, arriving just in time. But you can refuse the shipment. Stop performing devotion to the next imminent transformation. Use the tool or do not. Build things with it. Let it take you from zero to one. Let it handle the autocomplete while you do the thinking. The world will keep not ending either way, and the businesses that win will be the ones that spent that attention on real problems instead of on the drama.