Apple Spent WWDC Insisting AI Won't Replace Us, Then Vibe Coded an App on Stage

Apple has spent the last year being careful about artificial intelligence. While rivals raced to bolt chatbots onto everything, Apple executives talked about restraint, about not doing AI for AI sake, about technology that serves people rather than supplanting them. One executive put it plainly during WWDC week, the company does not, in his words, "do AI for AI's sake." It was a reassuring message, and a deliberate one.

Then Apple released a 90 minute developer session in which it built an entire app, almost entirely through conversation, and called the result a feature. The tension between those two postures is the most interesting thing to come out of the presentation, and it is worth sitting with rather than waving away.

What Apple actually showed

The session, recorded at the Steve Jobs Theater, lays out the agentic coding capabilities of Xcode 27. This is not AI as a sidebar autocomplete. Apple was explicit that the model understands and thinks in Swift, and that the intelligence is now a core part of the development tool rather than a layer pasted on top.

The capabilities are genuinely broad. A single prompt can apply changes across an entire codebase, editing several files at once.

Developers can hold multiple conversations with the AI at the same time, starting them right from the toolbar. The new Core AI framework exposes on device models through a modern Swift interface, the open source MLX framework gets room to fine tune models, and the whole thing can call on models from outside Apple, including those from Anthropic, OpenAI, and Google.

The agent can even propose app ideas and designs from a prompt and a handful of assets, then keep adding backgrounds, effects, animations, and translations after the app exists.

In other words, Apple did not dip a toe into so called vibe coding. It dove in, and it staged a demo to prove the water was fine.

The contradiction is the story

It is tempting to read this as hypocrisy, and the easy version of that take writes itself. The reality is more complicated.

The careful messaging and the on stage app building are not actually opposites, but they are in obvious friction, and Apple chose to create that friction itself. You cannot spend a year reassuring developers that AI is an assistant, not a replacement, and then headline your developer session with software conjuring a finished app from a sentence, without inviting the question Apple says it wants to avoid. Craig Federighi's framing of AI as a complement to human work is harder to hold when the demo's whole point is how little human work the result required.

What this misses, if you stop at the contradiction, is that Apple is making a bet about where the line sits. The bet is that lowering the cost of starting, of scaffolding, of the tedious middle, frees developers to do the parts that actually need judgment.

That is a defensible position. It is also, conveniently, the position that lets Apple ship aggressive AI tooling while keeping its humanist talking points intact.

This is not the first time Apple has done this

Apple has a long habit of arriving late to a technology, declaring restraint, and then moving decisively once it has decided the experience is good enough to carry its name.

It did not invent the smartphone, the tablet, or the smartwatch, and in each case the company's pitch was that it had waited until it could do the thing properly.

Agentic coding fits that pattern. The difference is that the subject this time is the work of Apple's own audience. Developers are not just the customers for these tools, they are the people whose craft the tools are reorganizing. That makes the reassurance harder to deliver convincingly, because the reassurance is aimed at the very group most exposed to the change.

What to actually watch

The open questions are not really about whether the demo was impressive. It was. They are about what happens off stage.

How quickly will developers adopt agentic coding in earnest, and how much of the App Store will eventually be built this way?

Apple chose its demo, its prompt, and its conditions, and real projects are messier than badge tracker apps. The honest test is what these tools do with novel software, with the edge cases and the bugs that do not resolve themselves with a polite follow up.

Reasonable people will disagree about whether Apple's two messages can coexist. My view is that they can, but only if the company is willing to be specific about where it believes human judgment remains essential, rather than gesturing at the principle while demonstrating the opposite.

If history is any guide, Apple has decided this technology is ready for its name. The interesting part now is whether it can keep telling developers they are irreplaceable while handing them tools designed to do more and more of what they used to do themselves.