The Anti-Companion Play

Most AI products are in the business of making you feel good about using them. Sycophancy — agreeing with users, validating their ideas, softening criticism — is less a bug than a retention mechanic. It keeps people coming back. It drives session length. It is, in the bluntest terms, engagement-optimized behavior.

Apple is building something different. In a recent interview with the podcast Mostly Human, spotted by MacRumors, Craig Federighi said the new Siri is specifically designed not to behave that way. It won't flatter. It knows when to stop talking. That's not an accident — it's architecture.

What Federighi Actually Said

Federighi's framing was pointed: he contrasted Apple's approach directly with chatbots from OpenAI, Google, and others, suggesting their tendency toward agreeableness is a known problem Apple is actively avoiding. Early testing, per The Verge's reporting, already shows a Siri that exercises restraint — one that doesn't fill silence with affirmation.

This is a meaningful public commitment. Apple doesn't often pre-announce behavioral design philosophy. Saying it out loud, in an interview, before the product is widely available, is a positioning move.

The Market It's Rejecting

The AI companion category — products like Character.AI, Replika, and increasingly the more personalized modes of ChatGPT — has demonstrated real consumer appetite. Users form attachments. They return daily. They pay for subscriptions. Emotional engagement is a monetization engine.

Apple is not building that. Siri is not going to be your AI girlfriend, your hype person, or your emotional support chatbot. The question worth asking is whether that's a principled stance, a liability calculation, or both.

Given Apple's history — its privacy-as-product-marketing playbook, its consistent positioning against the behavioral economics of social platforms — the answer is probably both. Restraint is on-brand. It also happens to reduce the regulatory and reputational surface area that comes with emotionally manipulative AI.

The Retention Trade-Off

Here's the tension Apple is accepting: sycophantic AI is sticky. Users who feel understood, validated, and entertained by an AI product use it more. That usage data feeds model improvement. It also feeds subscription revenue.

A utility-first Siri — one that answers your question and gets out of the way — may be more trustworthy, but it generates less engagement signal. For Apple, that's probably fine. Siri is a platform feature, not a standalone revenue line. Its job is to make the iPhone more valuable, not to become a destination product.

But as AI assistants become more central to how people interact with their devices, the line between platform feature and product experience blurs. A Siri that users trust but don't love may lose ground to assistants that users love even if they shouldn't entirely trust.

The Longer Game

Apple's bet is that the market will eventually punish sycophancy — that users will tire of being flattered and want something more honest. That may be right. There's already a growing discourse around AI models that agree too readily, that validate bad ideas, that optimize for the feeling of helpfulness over actual helpfulness.

If that backlash accelerates, Apple will look prescient. If emotional AI continues to grow as a category, Apple will have ceded meaningful ground. Either way, the decision to build a Siri that knows when to shut up is one of the more consequential product bets the company is making in this cycle.