The agent economy just got a price check, and it's not pretty.

The Summary

  • Manus AI, a desktop automation agent acquired by Meta for $2B+, can now organize your files, clean folders, and handle tedious computer tasks autonomously
  • The catch: aggressive usage metering makes routine tasks potentially more expensive than multiple software subscriptions combined
  • Manus joins Claude Cowork, Perplexity's Personal Computer, and OpenClaw in the emerging desktop agent space, each competing on capability while grappling with unsolved economics

The Signal

Meta didn't drop $2 billion on Manus for the feel-good story of organized Downloads folders. They're buying position in what might be computing's next interface layer, the agent that sits between you and your operating system. The desktop agent space is heating up fast, with multiple players racing to prove they can reliably control your computer without destroying everything you care about.

The functionality itself works. Sorting photos by event instead of month, tagging music metadata, organizing file hierarchies. These are real tasks that real people avoid because they're tedious but not quite painful enough to warrant existing solutions. That's the sweet spot agents are designed to own: valuable but boring work that humans resent doing themselves.

But here's where the model breaks. The metering problem isn't a bug, it's revealing something fundamental about how these agents actually work. Every action costs tokens. Every decision point burns through API calls. A task that takes you 30 minutes of mindless clicking might cost an agent hundreds of thousands of tokens in back-and-forth with whatever LLM is running underneath. The result: what feels like magic when it works becomes economically absurd when you look at the bill.

This is Web4's unit economics problem in miniature. We can build agents that genuinely save human time and attention. We just haven't figured out how to price them in a way that makes sense for anyone except the earliest adopters with expense accounts. The current model, aggressive metering with pay-per-token pricing, treats each interaction like a scarce resource when the value proposition depends on treating agent labor as abundant and cheap.

The Implication

If you're testing desktop agents now, track what you're actually paying per task. The winners in this space won't be the ones with the most impressive demos. They'll be whoever cracks sustainable pricing first, probably through flat monthly rates that absorb the token economics behind the scenes. Until then, Manus and its competitors are expensive proof of concept, not daily drivers. Watch for the first player to break ranks on pricing. That's when this category gets real.

Source: Fast Company Tech