Claude Code now spins up 100s of parallel agents on one task Anthropic's Claude Code now supports dynamic workflows that can spin up hundreds of parallel agents on a single task, enabling massive parallelism for complex software development. The update allows developers to deploy agent loops that coordinate multiple AI agents simultaneously, fundamentally changing how teams approach large-scale coding projects. unwind ai ../ - Posts - Claude Code now spins up 100s of parallel agents on one task Claude Code now spins up 100s of parallel agents on one task + Apple drops a native AI framework for on-deivce AI agents Today’s top AI Highlights: Apple Core AI Framework for on-device agents Printing Press: Print agent-native CLIs from a single prompt Claude Code Dynamic Workflows with massive parallelism Your job is to write agent loops now ChatGPT now "dreams" to build better memory & so much more Read time: 3 mins AI Tutorial The frontend used to be a fixed thing. Designers drew it. Engineers built it. Users got what shipped. That's over. The interfaces shipping in 2026 are drawn partly by the agent itself, in real time, from what the user actually asked for. Ask for a table, get a table. Not a paragraph describing one. Generative UI is the layer that lets agents stop describing and start showing. This guide walks you through three patterns that have emerged on how to build it, and the differences between them matter more than most teams realize. Latest Developments Forget calling external APIs. Apple's new Core AI framework, announced at WWDC 2026, gives developers Swift-native access to Apple's on-device foundation models with tool calling, structured generation, and full Apple Intelligence integration. This is the first time Apple has opened up its on-device models as a developer-facing framework. You write Swift, define tools, and the model runs locally on the device with zero cloud round-trips. Privacy-first by default, no API keys, no usage-based pricing, no latency from network calls. For anyone building iOS or macOS apps, this changes how you think about adding intelligence to your product. The bigger picture: Apple also revealed that Apple Intelligence is now co-developed with Google using Gemini models under the hood, running both on-device and through Private Cloud Compute. A system orchestrator automatically coordinates AI features across apps. Key Highlights: Tool calling in Swift : Define custom tools that the on-device model can invoke, enabling agentic workflows entirely on the user's device without a server. Structured generation : Get typed, schema-conforming outputs from the model, not just raw text. Build reliable features without post-processing hacks. Gemini under the hood : Apple Intelligence now runs on foundation models co-developed with Google, giving the platform multimodal capabilities, including image generation, visual Q&A, and speech generation. No cloud dependency : Models run locally. Your users' data stays on their devices. No API costs, rate limits, or cold starts. Available now : Core AI ships with iOS 27, macOS 27, and the latest Xcode. Documentation is live at developer.apple.com/documentation/coreai https://developer.apple.com/documentation/coreai?utm source=www.theunwindai.com&utm medium=referral&utm campaign=claude-code-now-spins-up-100s-of-parallel-agents-on-one-task . What if every app, API, and website your agent needs came as a purpose-built CLI with a local SQLite mirror, compound commands, and token-efficient output? That's Printing Press by Matt Van Horn and Trevin Chow. Point it at an API spec, a website URL, or even a service with no public API, and it generates a Go CLI, a Claude Code skill, an OpenClaw skill, and an MCP server. All from one prompt. Super interesting concept: a local SQLite mirror beats a remote API call. Compound commands beat ten round trips. An agent-native CLI beats raw HTTP. When you "print" an ESPN CLI, you don't get a thin wrapper around REST endpoints. You get live scores, series state, leading scorers, and injury news in one call, all queried from a local database that syncs incrementally. Same goes for Linear, Slack, Notion, or any of the 237+ CLIs in their Public Library. Key Highlights: No API needed : For services without a public API, Printing Press launches a browser, captures traffic, reverse-engineers the endpoints, and generates the spec automatically. If you can click through it, the press can build a CLI. Local-first data layer : High-gravity resources get domain-specific SQLite tables with FTS5 full-text search and incremental sync. Queries run in milliseconds offline. Your agent never waits for a 429. 237+ community CLIs : The Public Library ships pre-built CLIs across 19 categories, from flight search to restaurant reservations to eBay auctions. Install with one command or let your agent browse and pick what it needs. Token-efficient by default : --compact mode cuts 60-80% of tokens. Auto-JSON when piped. Typed exit codes for agent self-correction. The CLI is built for agents first, humans second. Try it now : Install via Go, add the Claude Code or OpenClaw skills, and run /printing-press