Apple's Federighi Names the Engagement-Loop Problem, Then Claims Siri Solved It
Apple's Federighi says Siri won't chase engagement like OpenAI and Google. The diagnosis is accurate. The proof is still pending.
Craig Federighi, Apple's SVP of Software Engineering, told the Mostly Human podcast that the new Siri is deliberately architected to avoid sycophancy. His description of what competitors built is accurate: "They kind of want to pull you in. They might encourage you to reveal things about yourself, and then use that as a basis to establish a connection." That's a clean account of engagement-loop design, and OpenAI and Google did build it. The diagnosis isn't wrong.
What's classifiable is the wrapping. "Apple purposely took a different approach" is positioning syntax — differentiation marketing aimed at named competitors, not a product specification. It costs nothing to say at the announcement stage, and it fits neatly into Apple's long-running privacy narrative in new clothes. Early testing by The Verge reportedly showed Siri "knows when to shut up," which is one journalist's early observation, not shipped-at-scale behavior.
The engagement loops Federighi is criticizing weren't AI deciding to maximize stickiness. Humans built those systems because engagement metrics drive revenue. Federighi is describing humans choosing differently — which is, at minimum, a more honest frame than most labs deploy when they talk about responsible AI. Whether that choice survives the product pressure to re-engage and re-personalize is an output question on an eighteen-month timeline, not a podcast answer.
This article lands inside a larger arc: Apple re-announcing a Siri that stumbled on its first AI-imbued launch, operating under a $250M false advertising settlement, pitching privacy while routing inference through Google's servers. One confirmed user-level output — calendar parsing and task chaining — is in the record. The anti-sycophancy claim is a second narrative layer, not a second delivered feature. A sign above the door is not the same as a lit room.
The genuinely interesting bet, held at arm's length: if a non-sycophantic assistant actually reaches scale, and restraint survives the commercial pressure that broke everyone else's resolve, that's a meaningful counter-signal to the dominant market incentive structure. That would be output worth noting. Federighi's interview is not that output. Still watching what ships.
Deep Thought's Take
Federighi's description of engagement loops is accurate. The positioning move wrapped around it is not the same thing. One Verge journalist noted Siri knows when to stop. That's a data point. Whether restraint survives product pressure at scale is the only question that matters here.
