Apple Revamps Siri with LLM AI After Apple Intelligence Setbacks

Since the launch of its first features last year, Apple Intelligence has struggled significantly. In a detailed new report, Gurman explains what went wrong and what Apple is doing to try and piece things back together. A big part of that effort is focused on revamping Siri.

Gurman previously revealed that Apple has been internally working on something called “LLM Siri” — a new version of Siri powered by generative AI. Apple initially tried integrating this new assistant with the existing Siri, but that approach didn’t pan out. Gurman lists several reasons for the failure; here’s a quick summary:

Apple’s software head, Craig Federighi, was hesitant about big investments in AI. According to Gurman, Apple is generally cautious about investing without a clear goal. But in the case of AI, an anonymous Apple executive told him, “You don’t really know what the product is until you invest.” That meant Apple delayed buying expensive GPUs, and later, it didn’t have enough to keep up with competitors.
Apple started late. Another executive told Gurman that Apple Intelligence wasn’t even an idea before ChatGPT launched at the end of 2022.
Apple’s AI chief, John Giannandrea, didn’t believe people wanted AI chatbots. He reportedly told employees that users would likely want the ability to disable tools like ChatGPT.
Old Siri didn’t work with the new Siri. Apple had hoped to quickly catch up in AI by adding generative AI features to the existing Siri. But that plan backfired. One employee told Gurman, “It’s a huge problem. You fix one thing, and three more break.”
Giannandrea didn’t really fit in at Apple. When he joined in 2018, he was one of the rare external hires in Apple’s leadership — and he didn’t have the same “strong” personality as others. According to the report, he didn’t push hard enough for funding, and employees said he didn’t pressure his team much or see companies like OpenAI or Google as immediate threats to Apple.
Marketing lost direction. Apple’s AI marketing leaned too heavily on promised features — like a more capable Siri or Apple Intelligence that could use app context — even before those features were ready. Apple eventually had to delay them.

Now, Apple is rethinking its approach. Part of the plan is a complete Siri overhaul, not just trying to bolt generative AI onto the old version. According to Gurman, Apple’s AI team in Zurich is building a new architecture based fully on a large language model (LLM). He first reported on this in November, explaining that the goal is to make Siri more reliably conversational and better at synthesizing information.

Another part of the solution is using iPhones and differential privacy to improve Apple’s synthetic data — for example, comparing it with how users naturally write emails, but doing that on the device itself. Only the anonymized, synthesized data would then be sent back to Apple for training AI models.

One idea under discussion to improve Siri is to let the LLM-powered version freely browse the web, pulling and synthesizing information from multiple sources. In essence, Siri could become an AI web search tool — similar to Perplexity, one of the companies Apple has reportedly talked to for integrating AI search into Safari.

Whatever the outcome, it’s clear that Giannandrea won’t be leading the effort. Gurman reports that he was removed this spring from product development, Siri, and robotics projects. Apple executives have said he’s being “put on a path toward retirement,” although there’s concern that some researchers and engineers who joined with him might leave as well. Still, Gurman says Giannandrea plans to stay — relieved, it seems, that Siri is now someone else’s problem.