My Laptop Works From Home while I am on a Business Trip to Zürich

March 25, 2026

Executive Summary

My laptop is executing tasks at home and I am steering it via voice command while sitting on the train to Zürich. I am not hallucinating. This is March 2026.

Q1 2026 did not announce a single breakthrough in the field of AI. It shipped a stack of products in under three months that, taken together, have fundamentally changed how we do our daily job. The most relevant are Cowork, Dispatch, Code Channels.

If you wonder what any of this means for how you will work every day in the future, this newsletter edition is for you.

IC 1, 07:30

This morning, I left the laptop at home and only took the phone with me to Zürich. However, the notebook is turned on, plugged in, connected to the internet and will execute tasks all over the day: Create a new power point presentation in corporate design, extract financial data, perform calculations, create excel sheet with charts… and so on but with no one sitting in front of it.

Fully autonomous without human control? Not at all. I can follow everything that the laptop is doing. I am frequently asked for permissions and I am giving instructions all day long. Typing or talking, whatever works best for the moment. It is like giving instructions to an intern or new team mate all day long, letting that person execute steps in front of the PC, while I am still supervising closely what is happening. A real person would maybe feel micromanaged after a while. My Laptop did not complain so far.

So how does this work?

Google on my phone converts my voice to text. That is not new. What is new: that text is not sent to a chatbot that replies and waits. It goes to an AI agent running on my laptop at home. The agent plans the task, executes it, and sends the result back to my phone. The laptop does the computing. The phone is the remote control.

That is Claude Dispatch, released on March 17, 2026. The permanent conversation between me on my phone and an AI agent on my laptop. This agent re-routes to the relevant application: Chat, Cowork, Coding. While I sit in a comfortable, punctual Swiss train enjoying a coffee, having nothing but my phone with me, I can continue working on my projects as if an employee were sitting in front of my laptop.

That is the new business reality.

We are not talking about the future any more. It happens now.

What Happened in Q1 2026

January 12: Cowork

Anthropic launched Cowork as a research preview, initially for Max subscribers, then rolled out to all paid plans. The concept: Claude gets access to a folder on your computer and executes complex, multi-step tasks autonomously from there. Creating documents, organizing files, compiling research, populating spreadsheets.

A chatbot answers a question. Cowork handles a task: it plans steps, coordinates sub-agents, checks in at decision points, and delivers a finished result. You describe the outcome, not the path.

The genesis is worth knowing. Claude Code, Anthropic’s terminal-based tool for developers, had become one of its most-used products since late 2024. Anthropic noticed something: users were forcing the coding tool to do non-coding work. Vacation research, slide decks, email management, recovering wedding photos from a hard drive. The underlying Claude agent was simply the best agent available, and people found ways to use it for everything. Cowork was the logical response: the same architecture, accessible without a terminal, without technical prerequisites. Anthropic reportedly built the entire feature in approximately a week and a half, largely using Claude Code itself.

November 2025 to February 2026: OpenClaw

To understand why Dispatch matters, a brief detour. OpenClaw was an open-source agent built by Austrian developer Peter Steinberger, controllable via WhatsApp, iMessage or Telegram, with unrestricted access to the entire local system. It became the fastest-growing open-source project of early 2026. Mac Minis sold out across Europe, purchased as always-on machines to run it around the clock. At Tencent’s headquarters in Shenzhen, Forbes reported that nearly 1,000 engineers queued to have it installed on their laptops.

OpenClaw proved the demand. It also proved the risk: Snyk found critical security vulnerabilities in roughly 7% of all available OpenClaw skills, and over 30,000 instances were reachable from the public internet. For regulated financial institutions, it was never a serious option.

Anthropic took note. Dispatch is, in essence, the same idea built properly.

March 17 and 20: Dispatch and Code Channels

Anthropic’s answer arrived in two steps.

Dispatch (March 17) connects the Claude mobile app to a running Cowork session on the desktop. The laptop computes. The phone steers. Instructions route through Anthropic’s infrastructure; execution happens locally on your own machine. Files never leave the computer. No API keys in Markdown files. Sub-agents can run tasks in parallel, visible in real time inside the Cowork application. Anthropic’s approach: sandboxed execution, controlled permissions, explicit approval before significant actions.

Code Channels (March 20) solves the same problem for developers using Claude Code: native Telegram and Discord integration. A developer types a message in Telegram, Claude Code executes the task on the local machine and replies in the same thread. No self-hosted relay, no webhook configuration, no third-party bridge software.

What This Means for Wealth Managers

The Agent as Staff

Previous editions of this newsletter have described AI as a tool: a capable instrument that responds to requests. Cowork and Dispatch shift that framing. The agent no longer waits to be asked. It handles tasks in the background, coordinates sub-tasks, delivers results, and can be directed from anywhere.

For a family office or boutique wealth manager, this translates directly: routine processes such as summarizing fund news, extracting structured data from PDF reports, or organizing meeting notes can be configured as recurring tasks. The agent executes them while the portfolio manager is with a client. This is the product offering as of March 2026.

The Security Question

OpenClaw answered one question definitively: the demand for autonomous agents is real and large. It also showed what happens when that demand is met by uncontrolled open-source solutions. Dispatch is explicitly flagged as a research preview, and Anthropic itself advises against using it for regulated workloads. That caveat is important and should be taken seriously.

The architectural direction is nonetheless unambiguous: local execution, controlled permissions, explicit approvals, auditable boundaries.

The New Competency Line

Cowork and Dispatch require a competency that has nothing to do with coding. Staff need to formulate tasks precisely enough for an agent to execute them autonomously, and evaluate the output critically. This is the same shift Excel triggered in the 1990s. Not every analyst needed to program. But those who did not understand Excel lost ground.

What This Means for Your Practice

Curious what this looks like in your specific environment? What is actually possible today, and what still needs to wait? If you are looking for someone who understands both the technology and the business context of wealth management, I am happy to schedule a call.

Reach out via LinkedIn or book directly at Gerevest.AI or via Calendly.

Sources:

About the Author: Dr. Andreas K. Janoschek specializes in AI applications for Asset & Wealth Management. Based in Geneva, he helps industry professionals stay ahead of competition by securely advancing with AI. Should you wish to discuss your specific situation and implementation approach, please do not hesitate to contact us.

This newsletter aims to inform and does not constitute investment or legal advice. Always consult with qualified professionals for specific circumstances.