Xcode 27 is a real departure from what Xcode has been for most of its life a code editor that happened to have some analysis tools bolted on. What Apple shipped at WWDC26 is closer to an agentic IDE that does localization, performance diagnosis, and project scaffolding on your behalf, while still being the place where you read and write Swift.
The workspace redesign is thorough enough to matter even if you never touch a coding agent. The toolbar is now fully customizable drag, reorder, remove items as you like and history navigation, branch picker, and editor mode controls have been repositioned so the layout is actually coherent. The new theme system is the more interesting part: a pair of sliders controls color intensity and background saturation, you can push the background all the way to a vibrant gradient, lock individual colors when you want them pinned, and assign a different theme per workspace so you can tell at a glance which project you're in. Xcode 27 is also now Apple Silicon exclusive, which cut the app's download size by 30%.
Inline issue indicators get a new "predictive" look — a subtle background tint instead of hard warning colors while you're actively typing, then full intensity if the issue persists through a build. It's a tiny change but it reduces the visual noise tax you pay just for editing code mid-thought.
Coding Agent Integration
The coding agent experience has been rebuilt around the editor itself, not a sidebar afterthought. Conversations appear as editor panes you can split and tab like any other file, and a new /plan command lets you have the agent gather context and write out a plan before touching any code, with inline feedback before it starts implementing. As it works, changed files and artifacts, including simulator screenshots from Previews and Device Hub, appear in a live panel alongside the conversation.
Sub-agents run in parallel for large tasks, and the agent sidebar shows all active conversations with unread message indicators so nothing falls through. The model lineup now includes Anthropic Claude and OpenAI Codex directly in the workflow, and GitHub and Figma are among the first to offer native integrations. MCP (Model Context Protocol) and Agent Client Protocol let you bring your own tools and compatible agents in as well.
Device Hub
Simulators used to just open in whatever window they felt like. Device Hub gives them a proper home: a compact, device-sized window with quick actions (home, screenshot, rotate), an expandable inspector for accessibility settings like contrast, Dynamic Type size, and dark mode, and the ability to test iPhone apps in the new resizable iPhone Mirroring context from macOS 27. Physical devices live in the same sidebar alongside simulators, so you can switch between an iPhone 17 Pro simulator and a paired iPad Pro without leaving Device Hub at all.
Localization by Agent
This is the most practically useful part of the whole session. You ask the coding agent to localize the app, it audits the code to make sure string literals are extraction-ready, creates String Catalogs, then fans translation work out to sub-agents in batches. Each sub-agent gets per-string context, cross-string terminology consistency from the catalog, and Apple's per-language style guides. A TRANSLATION.md glossary file lets you define domain terms, tone notes, and do-not-translate strings that carry over automatically when you add new languages. For subsequent languages, a "Generate Translations" button in the String Catalog UI kicks off the same flow without a new conversation.
Organizer Overhaul
The Organizer got four meaningful upgrades, not cosmetic ones. The new Overview page puts metric charts and diagnostic reports on one screen so you can correlate a crash spike with the exact version where it appears without switching views. There's a new Storage metric that breaks down documents, data, and binary size across releases useful both for cellular download size and for catching apps that silently accumulate cache bloat. The old Scrolling hitches metric has been replaced by a broader Hitches metric that catches animation problems throughout the app, not just in scroll views, including Liquid Glass and SwiftUI view transitions.
Metric Goals replace last year's launch-time-only recommendations with a calibrated target set covering hang rate, disk writes, battery usage, storage, and hitches, benchmarked against technically and functionally similar apps, plus your own historical baseline. And a "Generate Recommendations" button ties directly into the coding agent, so you can go from "the Organizer found a hang regression" to "the agent is reading the diagnostic data and proposing a fix" without context-switching to a separate tool.
Instruments Top Functions
Instruments adds Top Functions, a mode that surfaces the most expensive functions across the entire call stack, regardless of where they're called from, so you don't have to manually unroll a flame graph to find a hot loop hiding eight frames deep. Run Comparison landed in 26.4 and carries into 27 select two runs, see which functions got faster or slower. The Swift Concurrency instrument also gained dedicated Main Actor and Global Concurrent Executor tracks for visualizing task scheduling delays and actor contention, which is going to be genuinely useful for anyone building anything with Foundation Models.
Xcode Cloud setup is now a three-click flow from a "Get Started" button in the editor, and once connected, it runs builds and tests against every commit on main or feature branches automatically. That part hasn't changed philosophically, but making it frictionless enough that there's no excuse to skip CI is a real quality-of-life win for smaller teams who kept putting it off.
The throughline across all of it: Apple is betting that the right level for AI assistance is deeply embedded in your actual workflow not a floating chat box, but something integrated enough that the agent sees your simulator, reads your diagnostics, and edits your String Catalog while you're still in the same window. Whether that actually compounds into productivity gains or just moves the cognitive overhead around remains to be seen. But the architecture is there now.
#WWDC26