Microsoft has a vision for the future of Windows, and it is agentic. At its Build keynote back in June, the company showed off a native OpenClaw companion app for its operating system as living proof that the OS of tomorrow is one where autonomous AI agents live inside the desktop, hook into files, browsers and messaging apps, and just get things done. It was a bold statement. Then PCMag's Chris Hoffman actually installed the thing and spent days with it. His verdict, published August 6: the experimental, open-source tool is nowhere near ready for the limelight.
I think he is right, and more importantly, I think the gap between Microsoft's pitch and the shipped product tells us something important about where the agentic OS is going to stumble first: not on ambition, but on onboarding.
The Build Promise vs. the Install Reality
Let's give Microsoft credit where it is due. The company did not just bolt OpenClaw onto Windows; it built a proper graphical installer that sets up the underlying software with minimal user input. That is a real step forward for a project whose manual setup has historically been a wall of text and terminal commands.
But the ease ends almost immediately. The first real task, connecting a model provider, is where the consumer-friendly facade falls apart. You have options, technically:
- Use your ChatGPT subscription, the one vaguely mainstream path
- Connect an API key and pay per token through a service like OpenRouter
- Run models locally via LM Studio, which demands serious hardware
I am not sure that three options, two of which require you to think like a DevOps engineer, count as consumer choice. Hoffman used OpenRouter, and from the start it was obvious who this app is actually for: AI enthusiasts and developers comfortable with API endpoints, model choices and the low-level details that consumer chatbots hide.
Then there is the WSL elephant in the room. The OpenClaw Windows app is a friendly wrapper, but the engine underneath is still Linux software running through the Windows Subsystem for Linux. That is the kind of implementation detail that makes regular users' eyes glaze over, and it is the kind of thing that decides whether an easy install actually stays easy.
The Messaging Setup That Forgets You
OpenClaw's magic trick has always been that your agent lives in your messaging apps: you chat with it on Telegram or WhatsApp and it acts on your behalf. That is genuinely compelling. But getting there on Windows is the weakest part of the experience.
To text your agent when you are away from the PC, you have to enable at least one chat channel. The recommended Telegram bot is simple to configure, except that setup requires opening a terminal window and pasting a command. The Windows app does not fully replace the terminal for OpenClaw in general, something less technical users will likely prefer to avoid.
I cannot stress how different this is from what Apple is doing. With macOS Golden Gate, Apple's AI tools simply work without much configuration, living in approachable context menus and interfaces. Even ChatGPT and Claude's PC-focused agentic features are much easier to set up and use. The contrast could not be starker:
- The terminal never fully goes away
- You make model-provider decisions before you have run a single task
- Agents are increasingly expensive to run as API costs climb
- The reward, autonomy, is exactly what scares newcomers
The Verdict
Look, I want the agentic OS to work. The idea that your computer has a tireless assistant that can manage email, dig through files and coordinate your tools while you sleep is one of the few genuinely new ideas in personal computing in a decade.
But Microsoft is betting that the future of Windows is OpenClaw, and the current reality is a tool that remains complex and nonintuitive even when wrapped in a first-party app. It is a developer platform wearing a consumer suit.
Hoffman's bottom line is that OpenClaw is not anywhere near ready for the masses. I would add one nuance: it does not need to be ready for the masses today, but Microsoft needs to stop marketing it like it is.
The company is likely targeting developers and experienced enthusiasts with the app, which is a reasonable strategy. The problem is the framing. When your CEO tells the world the agentic future of the OS is here, and the first hands-on reports describe a maze of terminals, rising API costs and a tool that leaves regular users completely behind, you have a messaging problem as much as a product problem.
Microsoft has a lot of work to do if it ever intends for regular users to get on board. The good news is that OpenClaw itself keeps getting smaller, safer and more infrastructure-grade: the OpenClaw Foundation has acknowledged its rough releases and is restructuring core, plugins and release hygiene. The engine is improving.
The OS integration just is not there yet. And until setup stops feeling like a hobbyist project, the future of Windows will remain, for most people, a promise rather than a product.
Comments