WWDC 2026 Recap

Our Day 1 recap covered the consumer-facing changes: Siri AI, iOS 27, the Liquid Glass adjustment. The sessions that ran through the rest of the week carry the detail that matters for anyone deciding where to put engineering effort over the next two quarters. For practitioners shipping on Apple platforms, for consultants advising clients on AI build-versus-buy, and for the executives funding those roadmaps, here is the practical read on four of them.

Xcode 27 ships agentic coding, and it's model-agnostic

Xcode 27 integrates coding agents from Anthropic, Google, and OpenAI directly into the development workflow. Conversations support interactive planning, multi-turn Q&A, and a canvas that renders Markdown alongside code changes and previews. Agents can run autonomously by writing and executing tests, using Playgrounds for isolated testing, checking visual output through previews, and interacting with the simulator through a new Device Hub. Inline autocomplete runs on the Neural Engine, so source code stays on the Mac for a completion. Heavier work routes to the third-party models as an explicit opt-in, since that context leaves the machine. Xcode is now Apple Silicon-only, 30% smaller, and faster, with Xcode Cloud up to 2x faster. Pokde + 2

The architecture is the part worth understanding: private autocomplete on-device, frontier capability in the cloud on request. Xcode also supports extensions through plug-ins using the Model Context Protocol, with GitHub and Figma among the first to ship integrations. For consultants, that MCP support is the detail to track, since it means a client's existing agent tooling can connect to Xcode rather than being replaced by it. Pokde

Where to start:

• Pilot the agentic workflow on a contained task first, such as writing test coverage for an existing module, before letting it plan multi-file features.

• Set a policy on cloud agents up front, since that context leaves the machine while on-device completion does not. This is the governance question executives will want answered before the team adopts it broadly.

• If your team or your client already runs Claude Code, Cursor, or Copilot, evaluate Xcode 27 against them and check whether the MCP plug-in path keeps the current tools inside the IDE.

The Foundation Models framework and the cost of adding AI features

The headline from the Platforms State of the Union was free access to Apple Foundation Models running on Private Cloud Compute for developers with fewer than two million first-time App Store downloads. The framework also gains image input, server-side model integration that lets you call third-party models like Claude and Gemini through the same Swift API, and a new Dynamic Profiles system for multi-agent workflows. A new Evaluations framework lets you verify that AI features behave correctly across dynamic conditions, beyond what unit tests catch. Apple confirmed the framework will be open sourced later this summer. MacRumors + 2

For a head of AI weighing in-house inference against platform models, this changes the cost calculation directly. You can switch between Apple, Claude, Gemini, and other compatible models with minimal code changes, and image input lets apps analyze visual content, extract text, and read screenshots. The single-API abstraction also lowers the switching cost between providers, which is worth weighing against the lock-in of building on any one vendor. Tech Research Online

Where to start:

• Check your first-time download count against the two million threshold to confirm whether Private Cloud Compute access carries no cloud API cost for you.

• Prototype one backlog AI feature on the on-device model, then route to a cloud model through the same API where quality requires it.

• Build an evaluation set with the Evaluations framework before shipping, since model behavior varies across conditions that unit tests do not cover.

The capabilities framework for building in-app agents

This session covered the primitives behind running Apple Intelligence inside an app. Multimodal prompts let you pass images alongside text so the app can reason about visual content, with Vision framework tools like OCR and barcode reading available on-device. Dynamic Profiles let you swap models, tools, and instructions on the fly, so the app's behavior adapts within a single continuous session. A new View Annotations API maps your views to entities for on-screen awareness, so a user can reference and act on what is in front of them conversationally. The App Intents Testing framework validates Siri, Shortcuts, and Spotlight integration through real system pathways without UI automation. Apple DeveloperApple Developer

Dynamic Profiles is the primitive to understand. A single session can shift tools and instructions mid-flow, which lowers the amount of custom orchestration code required to run a multi-step assistant inside an app. For consultants scoping agentic features for clients, that reduction in glue code is a direct line item in the estimate.

Where to start:

• For a workflow with distinct stages, map each stage to a Dynamic Profile so the assistant changes tools and instructions as the user moves through it.

• Use View Annotations to expose key screens to on-screen awareness, beginning with the views where users most often ask what something is or want to act on it.

• Adopt the App Intents Testing framework early to validate the Siri and Spotlight paths without building UI automation around them.

What's new in Xcode 27 beyond the agents

The agents drew the headlines, and the session also covered changes that affect daily work regardless of AI use. Xcode 27 is Apple Silicon-only, which dropped Intel x86 support and shrank the download by 30%, and it adds a customizable toolbar and a new theme system. Xcode Cloud is up to 2x faster, and the new Device Hub unifies physical device and simulator management. SwiftUI and UIKit gained adaptive layout APIs for hinge-state detection and multi-configuration display handling, and Apple stated that fluid reflow rather than letterboxing is the expected default on flexible-screen hardware. Pokde + 2

The quieter item is the migration path. SiriKit apps keep working but will show deprecation warnings, and App Intents is the way Siri connects to an app, so an App Intents migration belongs on the plan. For executives, this is a budget and timeline question more than a technical one, since the work lands whether or not the team adopts anything else announced this week. LushBinary

Where to start:

• Confirm build machines and CI are Apple Silicon, since Intel support is gone.

• Place a SiriKit to App Intents migration on the roadmap rather than waiting for the deprecation window to close.

• If flexible-screen or foldable hardware is on the horizon for the product, adopt the new adaptive layout APIs and design for reflow rather than fixed breakpoints.

Across the four sessions, the through-line is consistent. Apple is lowering the infrastructure cost of AI features and positioning App Intents as the connective layer between an app and the system. For practitioners that sets the build agenda, for consultants it reshapes the estimate, and for executives it defines what the platform now provides without a separate vendor. Addressing both before the September release is the practical move.