Anthropic Just Gave You 3 Tools That Work While You're Gone.
TL;DR
Anthropic’s new trio turns Claude into a real “while you’re gone” worker — Nate argues scheduled tasks, Dispatch, and computer use finally create a secure, widely available agent stack that can deliver finished work, not just more summaries and briefs to read.
Dispatch is more than persistent chat on your phone — after pairing Claude mobile with desktop via QR code, your phone becomes a command surface that can spawn multiple parallel co-work sessions on your computer, each with its own context, files, and connectors.
Scheduled tasks close the “important but never urgent” gap — Anthropic runs jobs in its own cloud environment with configurable network access and setup scripts, and Nate highlights Anthropic’s own use case of keeping a Go/Python library in sync automatically.
Computer use matters because most business tools still have no clean API or MCP server — Nate calls MCP the “universal USB of the AI age,” but says old Jira setups, bespoke ERPs, SAP screens, and other antique systems are exactly where keyboard-and-mouse automation starts taking real work off your desk.
The practical benchmark is simple: does the agent remove work from your plate? — Nate dismisses “proactive briefing” demos and overused travel-planning showcases as pseudo-work, contrasting them with useful jobs like AI news monitoring, airfare tracking, bill reminders, research prep, and overnight code maintenance.
The bigger shift is psychological, not technical — he says 2026 is about learning to act like a manager, trusting agents to work while you’re away, and getting comfortable not hovering while Claude does hours of parallel work from your pocket.
The Breakdown
The real unlock: work that leaves your desk
Nate opens with a blunt distinction: most AI agent demos are optimized to look impressive, not to actually remove tasks from your life. His standard is simple — if the agent just produces another doc for you to review, that’s not leverage; the win is when the work is simply done while you’re out living your life.
Scheduled tasks as the first real primitive
He starts with Claude’s scheduled tasks, framing them as the first Lego brick: a cloud-run computer you can message with a repo, schedule, and prompt, even when your laptop is off. Anthropic apparently uses it internally to keep a Go/Python library in sync — exactly the kind of important, never-urgent maintenance work that usually slips through the cracks.
What this looks like outside engineering
Nate gets concrete fast: morning AI-news parsing, hourly airfare checks, and reminders for bills that still can’t be put on autopay. His point is that once native scheduling exists, a surprising amount of life starts to look “agent-shaped,” because anything tied to time and repetition can be handed off.
Dispatch turns your phone into a management console
The next piece is Dispatch, which he says has been misunderstood as mere persistent chat. In his telling, it’s really an orchestration layer: pair phone and desktop by QR code, then launch multiple independent Claude work sessions in parallel, with your phone acting as the command surface and your desktop as the execution surface.
The bounce-house example makes it click
His favorite proof point is product manager Pavle Hurin, who reportedly ran Dispatch for 48 hours while parenting, directing work from the sidelines of a kid’s bounce house. Hurin spent about 25 minutes entering commands while Claude handled hours of parallel competitor analysis and stakeholder messaging — a vivid example of what Nate calls “parallel asynchronization from your pocket.”
The caveats are real, but so is the shift
Nate doesn’t hide the rough edges: your desktop has to stay on, each subtask requests folder access separately, you can’t attach files from the phone yet, and complex multi-app jobs seem to succeed only about half the time in early testing. But he sees that as normal research-preview friction, not a reason to miss the broader shift into a true manager-agent workflow.
Computer use is the biggest deal for builders
For Nate, the most consequential release is computer use, because MCP connectors won’t cover the whole software world any time soon. He calls MCP the “universal USB of the AI age,” but says the real office pain lives in the unreachable long tail: old Jira instances, custom ERP screens, ancient SAP setups, and ugly manual workflows where someone still copies data between systems by hand.
Managed agent vs. self-hosted agent
He closes by contrasting Anthropic with OpenClaw: not safe vs. unsafe, but managed vs. self-hosted. OpenClaw gives you raw freedom and more control if you want to run the closet-server version yourself; Anthropic abstracts away the infrastructure, much like Gmail after self-hosted email or AWS after rack servers. The deeper challenge, he says, is human: learning to trust that the computer is actually doing useful work when we walk away — and building the management habits to take advantage of that.