Anthropic just shipped Claude Dispatch, and it’s basically the “text-your-agent-from-your-phone” dream… without the sketchy security and wallet-burning that comes with a lot of DIY stacks like OpenClaw. In this video I show a totally hands-off demo where I kick off real skills from my phone—scraping 200 agency owners in California, running my inbox cleaner, and even recreating a YouTube thumbnail with my face—while everything actually runs locally on my computer through Claude Co-work. The second I send a message on mobile, it reflects on desktop, spins up a Dispatch tab, runs the SOP inside the skill, and reports results back to my phone. The big takeaway: most autonomous agent tools are broken in practice—especially on security. We’ve seen tons of leaked API keys and basic mistakes, plus the “context explosion” cost problem where you’re paying API pricing per message/skill run instead of a subsidized subscription. Dispatch fixes this by routing through the Claude app, keeping things sandboxed/local, and using allow-list style permissions (it literally prompts you to approve access when a skill needs something). I also walk you through setup in a few minutes: update Claude Desktop + mobile, enable Dispatch in Co-work, pair your phone, keep your computer awake, and you’re live.