It Has to Be Great—Even Without AI

Steven Levy argued in Wired that Apple’s next CEO has to ship a “killer AI product.” John Gruber responded that AI is not a product category. It is pervasive technology — like wireless networking. Apple doesn’t need a killer AI product any more than it needed a killer wireless networking product.

Gruber is right. The implication for every CEO weighing AI strategy is sharper than his Apple-specific argument lets on.

If AI is a pervasive capability rather than a product category, the product still has to be great on its own. AI doesn’t excuse a weak experience. It enables new capacities that were not possible before. The bar for “is this worth using” is unchanged.

That is a different test than the one most enterprise AI strategies are passing.

AI is not a value proposition

The phrase “AI-powered” survived the bubble but is wearing thin. In 2023 it could move enterprise procurement. In 2026 it cannot. Buyers have been burned enough to ask the harder question first: what does this thing do, and is what it does worth paying for?

If the answer depends on the AI label, the product is broken. If the answer is “this does X well, and AI is part of how,” the product might be worth buying.

The wireless analogy is the right one. When Apple shipped AirDrop, the value proposition was not “wireless networking.” It was “send a file to the person across the table in two taps.”

Wireless is what made it possible. AirDrop is what made it a product.

The same will be true of AI. The products that use AI well will be products whose value can be explained without using the word “AI.” AI is what makes them possible. It is not what makes them worth buying.

The test

For any AI product pitch a CEO hears:

Strip the word “AI” from the pitch. Read what is left. Does it sound like a product someone would still buy?

If yes, the AI is part of what makes the product good. The value sits in the product, not in the technology. Build it.

If no, the AI is doing the marketing. The pitch is a model wearing a product costume. Don’t build it.

The test separates two kinds of AI products. “AI as capability” — where the product is great and AI is part of how it is great. “AI as crutch” — where the AI label is propping up a product that does not otherwise stand.

The second category will not survive the decade. Buyers are already filtering for the first. Where the AI ends up doing real work, the competitive position lives further upstream — in the data, the objective, or the human judgment the algorithm sits on top of.

What AI enables

AI is the technology that makes prediction cheap¹. When prediction gets cheap, you use more of it — and you use it where it was previously too expensive.

Three capacities follow. Each is a place where cheap prediction unlocks an experience that was not possible before.

1. Natural language interfaces.
   Predicting what a user means from what they say. Software controlled by speaking, not clicking — not “chat” as a feature, but chat as a substitute for navigation.
2. Personalization at scale.
   Predicting what each user wants without a content team writing rules. Each user sees a different surface, automatically.
3. Pattern recognition in unstructured data.
   Predicting structure where none was explicit. The system reads pdfs, emails, voice notes, and operates on them as if they were structured.

None of these are products. All of them make some products possible that were not possible before.

A CEO’s job in 2026 is to find the match. Which of these capacities, applied to which existing products or services, enables an experience the company could not previously deliver? Then ship that experience as a great product, not as an AI announcement.

• • •

The companies that win the next decade of AI will be the ones whose products survive the strip-AI-from-the-pitch test. The losers will be the ones whose decks are full of “AI-powered” features that disappear when you cover the prefix.

AI is technology. Products are still products. The bar has not moved.