Perplexity Teammates: always-on AI agents for $200/mo Max users
Persistent agents that run jobs in parallel and ping you in Slack — plus Google's 10M-token Muir, outbound phone agents in the OpenAI API, and Siri opening up.

Copy markdown
Teammates that run in parallel
Perplexity's new Teammates are always-on agents you assign jobs to — each runs in its own thread, works simultaneously, and reports back through Slack or the on-device Perplexity Computer. It takes the $200/mo Max plan, pricier than OpenAI's Dots or Anthropic's Claude Duet. Live for Max users on web now; mobile and desktop next.
Muir keeps 10M tokens in 25MB
Google researchers' Muir drops the growing KV cache for a fixed-size recurrent state: a 1.34B model trained at 4K context extrapolates to 10M-token streams with near-100% passkey retrieval to 1M, while that state stays ~25MB against a Transformer's reported 2.1TB. Training runs 2.7x faster at 32K. No weights yet — it's a paper — but a clear signal of where cheap long context is heading.
Your GPT-Live agent can dial out
OpenAI switched on outbound SIP calling for every org on the Live API — point GPT-Live at a phone number through your SIP provider and your voice agent can place calls, not just answer them. Appointment-setters, reminder bots, and follow-up callers are now a weekend build.
Siri opens to any app's App Intents
Apple's iOS 27.3 betas let Siri invoke App Intents from any app, not just a hand-picked few — a developer preview that finally makes Siri a usable voice-action layer. Expose your intents and users can trigger your app's actions by voice.