Apple's Siri AI Fix Feels Flat — But Xcode 27's Coding Agents Don't

Siri Finally Works — and Almost Nobody Cares

Apple has, at long last, shipped the Siri it promised back in 2024. The rebuilt assistant — now branded "Siri AI" — landed in the iOS 27 consumer beta in July, following its developer debut at WWDC on June 8, with a full public release expected in September115. It can hold natural, multi-turn conversations, understand personal context across Mail, Messages, Photos, Files and Calendar, react to what's on your screen, pull live answers from the web, and — for the first time — lives in a dedicated app with synced conversation history1218.

By nearly every measure, it's the assistant users were promised two years ago. And that, precisely, is the problem. The dominant reaction across the coverage has been a shrug: Apple didn't revolutionize anything, it fixed a bug — the forever-broken Siri — and delivered in 2026 what should have shipped in 202413. Being a functional AI assistant "isn't the breakthrough that it once would have been," as one assessment put it, because AI tools are already writing software, running multi-step agent workflows, and reasoning alongside users on their own machines25.

The Irony: The Most Genuinely Advanced AI Apple Shipped Wasn't Siri

Here's the part of the story that the anticlimax framing undersells. While the Siri overhaul is a consumer catch-up play built partly on Google's Gemini models118, Apple quietly shipped something far more current in the same release cycle: Xcode 27 treats AI coding agents as first-class citizens of the IDE. Where Siri AI feels like Apple arriving late to a party that started years ago, the agentic coding toolchain puts Apple ahead of, or at least level with, where the industry actually is right now1221.

Xcode 27 ships with coding agents from Anthropic, Google, and OpenAI built directly into the editor, with interactive planning, multiturn Q&A, and a canvas that renders Markdown and code previews alongside the conversation2421. It adds a two-tier AI architecture: on-device inline completions running entirely on Apple Silicon's Neural Engine — no source code ever leaves the machine — with cloud agents from Claude, Gemini, and Codex available for heavy lifting like multi-file refactoring and full test-suite generation, behind explicit opt-in2327.

The deeper structural story is protocols. Apple adopted the Model Context Protocol, the open standard that Anthropic introduced in late 2024 and that OpenAI, Google DeepMind, and now Apple have all embraced, via a binary called mcpbridge that translates MCP calls over XPC into Xcode's live process — giving external agents structured access to diagnostics, SwiftUI previews, the simulator, and the Swift REPL2330. Xcode 27 also introduced the Agent Client Protocol, which governs which agents can connect in the first place, meaning any third-party agent that implements ACP can plug in — not just the three launch providers1225. For a company famous for closed ecosystems, adopting the industry's open agent standards instead of inventing an Xcode-only extension system was, as one practitioner put it, "the least Apple-like decision of the keynote and the most welcome one"2623.

Apple Ate Its Own Dog Food — and That's Why Siri Got Fixed

The connective tissue between these two stories is one of the strangest corporate ironies of the year: Apple reportedly fixed its AI-laggard Siri team by forcing that team to adopt the very AI coding tools it was competing against. In April, reports emerged that Apple sent just under 200 Siri engineers to a multi-week bootcamp to learn agentic coding with tools like Claude Code and OpenAI's Codex1314. The Siri group had a "reputation as a laggard inside Apple" — other departments had already allocated large budgets to Claude Code while the assistant team hadn't kept up1314.

The restructure around the same time was brutal in its implications: roughly 60 engineers stayed on core Siri development, another 60 were assigned to evaluate the assistant's performance and safety, and the team was moved out from under former AI chief John Giannanderea — who stepped down in late 2025 and retired this spring — and placed under Craig Federighi, with Vision Pro veteran Mike Rockwell running Siri day to day1419. Rockwell would later stand on the WWDC stage and admit previous revamp attempts "didn't meet Apple's standards"20.

So the sequence reads almost like a fable: the team that couldn't build a modern AI assistant was sent to school to learn modern AI-assisted development, and two months later delivered a fundamentally rebuilt Siri1113. Apple turning agentic coding from an internal embarrassment into the default workflow of its flagship IDE in the same breath is the clearest signal yet that AI coding assistants have crossed from competitive advantage to baseline infrastructure1719.

Where the Coverage Agrees — and Where It Diverges

The reporting converges on the facts of the launch: iOS 27 beta in July, full release in September, Gemini-based models underpinning Apple Foundation Models, personal context and onscreen awareness as the headline capabilities11820. It also converges, remarkably, on the verdict: genuine improvement, muted reception.

It diverges on what the anticlimax means. One reading, common in the tech press, is that this is simply about timing — the AI race surged ahead while Apple dawdled, and a competent chatbot can no longer generate excitement28. Another reading, sharper in the trade coverage, treats the Siri launch as almost beside the point: the real WWDC 2026 story was the developer stack, where the Foundation Models framework expanded into a platform layer with multimodal input, server-model support, Python SDK availability, and connectivity to external models like Claude and Gemini — an unusually open posture for a company long allergic to lock-in-breaking abstractions712. A third strain of coverage was outright skeptical, noting user reports of glitches and arguing the update may be "too little, too late" against assistants that had already won user trust5.

My reading: the skeptical strain is the weakest. The beta-glitch complaints are noise; the structural shift is not. But the timing critique and the platform critique are both right, and they're the same critique from different angles. Apple's consumer AI is late because its developer AI arrived late — the Foundation Models Framework only opened to third parties at WWDC 2025, and Xcode's agentic capabilities only matured this year713. Siri AI's real moat isn't the model, which Google largely supplied; it's the orchestration layer — App Intents, Spotlight integration, Private Cloud Compute, and system-level permissions that external assistants can't reach1218.

AI Code Review Quietly Becomes the New Battleground

The agentic workflow logic that Xcode 27 bakes in — agents that plan, write code, run tests, watch previews, and revise their own output — is precisely the feedback loop that AI code review tools have been building separately1226. Apple's answer to that trend isn't a dedicated review product; it's structural. Agents validate their own work by running tests, trying ideas in Playgrounds, checking visual changes with previews, and interacting with simulators through the new Device Hub, so they "can run autonomously for longer"2422. That's a review function fused directly into the execution environment — something terminal-based agents like Claude Code have to duct-tape together with screenshot scripts and simulator MCP servers, and that Apple can do natively precisely because it controls the hardware2623.

The one-click integrations announced with Xcode 27 point the same direction: GitHub and Figma were the first partners to offer seamless installation between their tools and Xcode, meaning version-control data and design files become agent-accessible context2421. If code review in 2027 happens as an agent conversation over diffs, previews, and test results inside the IDE, Apple has positioned itself as the host environment for it rather than a participant competing on model quality.

The Anticlimax Is the Wrong Story

None of this rescues the Siri moment from feeling flat — a September release that lands two years after the promise, initially English-only, absent from EU iPhones and iPads over data-access regulatory disputes, and unavailable in China1218. Those constraints will mute the launch further.

But "anticlimactic" measures Siri against the wrong baseline. Measured against the 2011-era voice assistant it replaces, Siri AI is transformative; measured against the AI coding-agent ecosystem Apple simultaneously embraced — for its own engineers and for millions of developers — the Siri reboot looks less like Apple lagging and more like Apple deciding which race it actually intends to win. The consumer assistant needed to be adequate. The developer platform is where the company has quietly moved to the front of the field, and that's the part of this story that won't feel anticlimactic by next September71221.