What’s Happening
Manus AI — the Chinese-founded AI agent startup that made waves in early 2025 with its autonomous web-based agent — is pushing deeper into the personal computing stack. Its newest product, My Computer by Manus AI, is a desktop application that lets users automate files, applications, and multi-step workflows directly on their local machines. Think of it as the leap from a browser-based AI assistant to something that actually lives on your computer, can open your apps, move your files, fill out your spreadsheets, and chain together tasks you’d normally slog through manually.
The timing isn’t accidental. After the initial Manus AI launch drew comparisons to OpenAI’s operator tools and Anthropic’s computer use demo, the company has been iterating fast. The desktop client — currently rolling out in stages — signals that Manus is betting big on a future where AI agents don’t just answer questions in a chat window, but physically manipulate your operating system on your behalf. It’s a significant escalation in ambition, and it arrives as every major AI lab is racing to ship agent-style products. As TechCrunch reported during Manus’s initial launch, the company positions itself as a “general AI agent” capable of completing real-world tasks end-to-end — and this desktop move is the logical next step.
What makes this especially interesting is the modality shift. Cloud-based agents are useful, but they’re limited to what they can do inside a browser sandbox. A desktop agent that can interact with native apps — your IDE, your file system, Slack, Excel, Figma, whatever — is a fundamentally different product category. And it’s one that, until now, has been mostly theoretical outside of research demos from Anthropic and others.
Why Founders Should Care
If you’re building a startup, you’re probably drowning in operational busywork. Renaming files, updating project trackers, reformatting data, moving information between apps that don’t talk to each other. According to McKinsey’s research on generative AI, roughly 60-70% of workers’ time is spent on tasks that are at least partially automatable with current AI technology. Desktop automation agents like Manus’s My Computer product are targeting that exact gap — not with rigid scripts like traditional RPA, but with flexible, language-driven AI that can adapt to new instructions on the fly.
The market for AI-powered automation is enormous and growing. Grand View Research pegged the global intelligent process automation market at over $16 billion in 2025, with double-digit growth expected through the end of the decade. But here’s the real signal for founders: this isn’t just about enterprise workflows. Manus is going after individuals and small teams — the exact demographic that can’t afford a six-figure UiPath contract but desperately needs to claw back hours in the day. If desktop-native AI agents become reliable enough, they could become as foundational to startup productivity as Slack or Notion.
There’s also a platform play here. If Manus can establish itself as the default desktop AI agent layer — the thing that sits between you and your apps — it becomes incredibly sticky. The switching costs for something that knows your file structure, your workflows, and your habits are high. For founders evaluating whether to build on top of this kind of platform, or compete with it, the clock is ticking.
Who’s Already Moving
Manus AI (manus.im) is the obvious headliner. Founded by Yichao “Peak” Ji and his team, the company exploded onto the scene in early 2025 with a general-purpose AI agent that could browse the web, write code, and complete complex multi-step tasks. My Computer is their desktop-native evolution, designed to operate across macOS and Windows environments. Their approach leans heavily on large language models combined with computer vision to “see” and interact with the desktop UI.
Anthropic deserves mention here. Their “computer use” capability — demonstrated in late 2024 with Claude — showed an AI model controlling a desktop environment, clicking buttons, and navigating apps. While Anthropic hasn’t shipped a standalone desktop product, they’ve made the underlying model capability available through their API, and it’s the technical foundation that many startups (including, reportedly, some of Manus’s pipeline) draw inspiration from.
Adept AI, the San Francisco-based startup, has been working on its ACT-1 model — an AI specifically designed to take actions in software. They’ve raised over $400 million and are building toward a future where AI can use any software tool the way a human would. They’re more enterprise-focused, but the thesis is the same: software-using agents as the next platform layer.
Rabbit and Humane, while more hardware-centric, represent adjacent bets on AI agents that interact with the digital world on your behalf. Both stumbled with their initial hardware launches, but the underlying ambition — an AI that does things for you across apps and services — maps closely to what Manus is attempting through software alone.
“The next great platform shift won’t be a new device. It’ll be an AI layer that sits on top of all your existing tools and actually uses them for you.” — Yichao Ji, CEO of Manus AI, in a recent interview
The Startup Opportunity
So where does this leave founders who want to build, not just use, in this space? A few concrete angles:
- Vertical desktop agents. Manus is going horizontal — a general-purpose agent for anyone. There’s a massive opportunity to build desktop automation agents tuned for specific professions. Imagine a desktop agent purpose-built for accountants that automates QuickBooks, Excel, and email workflows in a way that understands tax codes. Or one for recruiters that operates across LinkedIn, your ATS, and Gmail seamlessly. Vertical specialization is where startups win against platform players.
- Agent infrastructure and middleware. As desktop agents proliferate, someone needs to build the plumbing. Think authentication layers that let agents safely access your apps, audit trails that log every action an agent takes (critical for regulated industries), or sandboxing tools that let you test an agent’s workflow before letting it loose on production data.
- Agent-native apps. Here’s a contrarian idea: instead of building agents that work with existing software, build software that’s designed from the ground up to be operated by AI agents. APIs are great, but most desktop apps weren’t designed for programmatic interaction. A new wave of “agent-first” tools — spreadsheets, design tools, project managers — could emerge that are optimized for both human and AI users.
- Trust and verification layers. One of the biggest barriers to adopting autonomous desktop agents is trust. Founders could build products that sit alongside agents and verify their actions — essentially, a “copilot for your copilot” that flags when an agent is about to do something risky, like deleting files or sending an email to the wrong person.
Watch Out For
Let’s be real: there are serious reasons this could go sideways.
Security is the elephant in the room. An AI agent with access to your desktop, your files, and your apps is a juicy attack surface. If Manus’s agent gets compromised — or simply hallucinates at the wrong moment — it could delete critical files, send sensitive data to the wrong recipient, or execute transactions you never intended. The history of automation is littered with examples of scripts gone wrong, and adding AI unpredictability to the mix raises the stakes considerably. Enterprises will be slow to adopt until there are bulletproof audit and rollback mechanisms.
There’s also the reliability problem. Current AI agents are impressive in demos but fragile in production. UI elements change, app updates break workflows, and edge cases multiply fast. Anyone who’s used Manus’s web agent knows it can nail a task on the first try and botch a nearly identical one the next. For desktop automation to become a daily driver and not just a party trick, reliability needs to go from 85% to 99%+ — and that last 15% is brutally hard to close.
Platform risk is real. Apple and Microsoft both have strong opinions about what software should be doing on their operating systems. Apple’s increasingly locked-down macOS permissions model could throttle desktop agents. Microsoft, meanwhile, is building its own Copilot agent layer directly into Windows. If Microsoft decides that third-party desktop agents are competitors rather than partners, it could make life very difficult for Manus and anyone building in this space.
Finally, there’s the regulatory uncertainty. The EU’s AI Act is already creating compliance headaches for AI companies. An autonomous agent that takes actions on a user’s computer — potentially making decisions with financial, legal, or personal consequences — is exactly the kind of high-risk application that regulators are paying attention to. Founders building here should be thinking about compliance from day one, not as an afterthought.
The Bottom Line
Manus AI’s My Computer product isn’t just another AI tool launch — it’s a signal that the desktop agent era is arriving faster than most founders expected. The shift from chatbots that talk to agents that do is the most consequential transition in productivity software since the smartphone. If Manus (or its competitors) can solve the reliability and security challenges, we’re looking at a fundamental rewiring of how knowledge work gets done.
If you’re a founder, here’s what to do with this: first, try the product. Get a feel for what desktop agents can actually do today versus what the demos promise. Second, look at your own workflows and your customers’ workflows — where are the repetitive, multi-app tasks that an agent could handle? That’s where the next wave of startups will be built. And third, if you’re building in this space, don’t try to out-general Manus or Microsoft. Go vertical, go deep, and own the trust layer. The horizontal agent war will be won by well-funded incumbents. The vertical agent opportunity? That’s wide open.
Leave a Reply