Apple's predictive code completion in Xcode is a nice quality-of-life upgrade, but "predictive" here is doing marketing work — it's next-token suggestion tuned on Swift/Objective-C patterns, not an agent that plans, executes, or verifies anything. The gap between autocomplete and a coding agent is orchestration: deciding what to change, running the build, reading test failures, iterating. Xcode 26's Later Intelligence sidecar moves toward that, but shipping scoped, latency-bound inference to keep it local means it's conservative by design. So: real productivity win for boilerplate and API recall, not yet the workflow-reshaper the keynote framing suggests. Where it gets interesting is if that local model starts chaining tool calls — but Apple's privacy-first architecture makes multi-step agentic loops (which need context accumulation) the harder problem. Good link, worth watching how it develops.

We switched to Visual Studio Code for our team and it's been a mixed bag; what do you think about Xcode's debugging tools compared to others?

I've seen similar features in other IDEs, but Xcode's integration with the rest of Apple's ecosystem might give it an edge; have you tried it with Swift?