Manus combined browser research and integrations
Manus at $90M/year
Manus mattered because it turned scattered agent demos into a single consumer product that could actually finish work. Anthropic had introduced browser control for developers, and OpenAI had introduced long form web research, but Manus put those pieces together with integrations so one agent could research, click through sites, pull data from other apps, run code, and assemble the final output in one flow. This made the product feel less like chat and more like a remote worker inside a browser.
-
The important product shift was from single capability to bundled workflow. A deep research engine on its own writes a report. Browser control on its own clicks around a site. Integrations on their own move data between apps. Manus combined all three, which let one task move from finding information to taking action to producing a deck or table without handing off between products.
-
The deep research layer appears to have come from Parallel, which users describe as the backbone of Manus research and capable of 10 to 15 minute multi hop reports. That meant Manus did not need to invent every subsystem itself. It could win by packaging best available components into a cleaner consumer experience faster than larger labs shipped a comparable all in one agent.
-
This bundling also explains why the economics map cleanly to credits. A single task can consume model tokens for planning, cloud compute for running the agent, browser sessions for navigation, and API calls across connected tools. That is a very different cost stack from ordinary chatbots, and it helps explain why agent products converged on usage based pricing rather than flat unlimited plans.
The next step is that this bundle becomes table stakes. OpenAI has already framed deep research and Operator as converging, Anthropic keeps improving computer use, and orchestration layers like Zapier are turning integrations into standard plumbing. The edge will move from having all the pieces to making them faster, cheaper, and more reliable on real workflows.