A First Look at the AI Hype · Cheap, Low-Trust, Low-Expertise AI-Made Work in an Oversupplied Market · The Price of Cost-Efficiency and Trust

The mood is beginning to shift away from simply calling for ever more token usage. Faced with budgets being consumed far faster than expected, major technology companies are starting to impose limits on how many tokens employees are allowed to use. The problem is that heavy token consumption is not necessarily translating into measurable gains—or, at least, not yet into gains that can be measured convincingly. This has given CFOs reason to reconsider how AI should be used and whether its economics actually make sense. That position differs from the rhetoric of major CEOs who had been urging aggressive and explosive adoption of AI. Of course, those statements were probably driven at least as much by business considerations as by wholehearted confidence in the technology itself.

Even from the perspective of ordinary consumers, we seem to be reaching the point where people are asking whether making money through AI is really as realistic as it first appeared. Software development is perhaps the most accessible example of AI-based monetization. Yet while AI can readily write code that works, it does not immediately produce a production-ready product. Reaching that stage still appears to require considerable trial and error, learning, and judgment from the user. To put it another way, AI can easily build a prototype. But stable mass production, packaging, distribution, and marketing still require a substantial amount of manual work, even when AI can assist with each of them. Every one of those stages requires domain knowledge. If a bottleneck appears anywhere along the chain, it becomes difficult to generate meaningful profit—profit that actually exceeds the cost of using AI.

AI can visibly reduce costs for almost anyone, across almost any area of interest. But creating genuinely competitive value still requires a great deal of manual intervention, relevant knowledge, and learning. Because reducing costs with AI is comparatively easy, it can also lead quite easily to reductions in headcount. Creating entirely new added value, however, is much harder. By contrast, individuals and organizations that already have a proven revenue model, demonstrated performance, and a deep understanding of their business are much more likely to produce results that continually justify spending money on AI. In other words, those who already possess an advantage in their domain—and especially those who already generate stable and substantial profits—may be the ones best positioned to use AI as leverage. That may eventually become a source of anxiety over a new kind of inequality. Most ordinary users will probably struggle to find a convincing economic justification for paying large amounts for AI. And the moment that realization becomes widespread and socially understood may become one of the temporary obstacles to the AI industry’s current boom.

When you build quickly with AI,

a product made quickly can attract attention at first, but ordinary users tend to lose interest just as quickly. There is often very little room to explore, experiment with, or play around with it.

Once a product has been released, fixing it can become much more complicated. Interesting apps usually accumulate a great deal of user data, and improving the product without damaging that data becomes an entirely new problem.

And if problems keep occurring, users eventually learn from the experience and start looking elsewhere.

“Build fast and ship first” is not a universal solution.

People often say, “It looks AI-generated.” What they usually mean is something closer to: “It looks cheap, and I don’t trust it because it looks like it was made with AI.”

People tend to overtrust their own AI-assisted work while being surprisingly quick to distrust the AI-assisted work of others.

The greatest weakness of today’s AI-made products as market goods is trust. And because this distrust is directed not necessarily at the substance of the product, but at the mere fact that AI was involved in making it, overcoming it will likely take a considerable amount of time.

Dominant platform companies are placing an increasingly high value on that trust.

Just as YouTube has taken increasingly aggressive measures to control the quality of AI-generated video, dominant platforms are steadily raising the barriers for producers who want access to their markets. Major app stores, for example, have all become considerably stricter about developer verification and app distribution.

In other words, today’s AI-made products are increasingly perceived as an oversupply of low-trust goods, while the marketplaces distributing them are using their superior position to create ever more elaborate filtering mechanisms.

The market does not readily trust what you made with AI.


+
From the user’s point of view, however, this may actually be good news.

When AI first emerged, people often spoke as though existing platforms were about to collapse almost immediately. Instead, those platforms are finding ways to survive by sharpening one of their most important advantages: trust. SaaS, for example, was repeatedly predicted to face imminent decline, yet it is now finding new ways to create value and appears more likely to coexist with AI than simply disappear because of it.

AI often gives you something close to a 100-out-of-100 answer. The problem is that in the real world, for reasons such as cost, you often do not actually need a 100-out-of-100 answer.

What you need instead is adjustment: finding the appropriate compromise between quality and cost. And AI is, in fact, potentially far better at this than humans. There is only one condition. It must have access to all the information about your intentions, your environment, your constraints, and your circumstances.

Unless AI stands in exactly the same position as you—seeing, hearing, and receiving everything you do—you are still needed as the person making the adjustments. Someone still has to understand the technical language in the AI’s answers and turn that understanding into precise instructions.

That is one reason you may continue to have a place in your profession. And it is also one reason your role will have to change.

If marketing is not your strength, it may actually be fine if users do not arrive quickly.

No matter how well you build it, the first version of your app will almost always be rough and incomplete. The users who actively search for an app like yours because they genuinely need it—the very people most likely to become loyal early users—may discover your app first, try it, and leave just as quickly.

And they learn something: “Yeah, that app wasn’t very good.”
They become former users who may not return easily.

Apps that urgently need rapid promotion are usually apps that continuously generate costs. The largest of those costs is labor. Labor costs do not fall easily, and they keep forcing an organization to spend money simply to remain alive. Survival therefore requires fast and capable marketing.

Your app, however? Its labor cost is “zero.”

If you are not particularly good at promotion, that weakness may buy you something valuable: enough time to improve the app while losing fewer of the people who might eventually have become loyal users.

So there is no need to rush.