Anthropic is on fire: here are three new features they just dropped (their desktop app looks seriously powerful now)
1. Desktop Claude Code (Epitaxy). There's now a terminal built right into Claude Code. It can work with multiple repositories at the same time. They also redesigned the layout: terminal, plan, current tasks, and preview are all neatly laid out in front of you, plus a bunch of handy shortcuts.
2. Claude Code Routines. Background code automations that run directly on Anthropic's infrastructure. Set up a rule (like "every night, pull bugs from the tracker and open draft pull requests"), and Claude works while you sleep. No need to keep your own computer on.
3. Hidden Assist for iOS. Enthusiasts dug up a new interface in the mobile app. It's not just an auto-clicker — it's a system for iteratively achieving long-running goals. You set a high-level KPI, and the model breaks it down into tasks and executes them in the background on its own.
It feels like all the labs right now are racing to make AI something you can hand off long-running work to, not just ping-pong prompts back and forth. Whoever gets this working reliably — and without the model losing track of the original goal by step ten — is going to hit the jackpot.
*And anyone who starts cutting limits is going to have their ass handed to them!*