Adept's UI-first agent strategy

Diving deeper into

/dev/agents

Company Report
Adept AI ($415M raised) is training AI to use existing software interfaces like a human would, rather than building a new OS for agents.
Analyzed 6 sources

Adept’s bet is that the fastest path to useful agents is to meet software where it already lives, on screens, in forms, and behind buttons, instead of asking every app vendor to expose new agent APIs. That makes the product look less like a new operating system and more like a trained digital worker that can read a web page, click the right field, type into it, and hand the final step to a person when judgment or approval is needed.

  • This approach matters because most enterprise work still happens inside old web apps and internal tools. Adept trains on real software usage and focuses on locating UI elements like buttons, links, and text fields, which lets it automate workflows without customers rebuilding their stack first.
  • The tradeoff versus an agent OS is control. An OS level system can see more context across devices and apps, but it needs deep platform access and distribution. Adept’s UI first model is easier to deploy into today’s software, but it wins only if it becomes highly reliable on messy real world interfaces.
  • The funding and partner list shows why this was strategically important early. Adept raised $350M in its March 14, 2023 Series B from investors including Microsoft, NVIDIA, and Workday, positioning it as infrastructure for enterprise automation rather than a single consumer assistant.

This category is moving toward hybrid systems that combine API level integrations where available with screen level automation where they are not. The winners will be the companies that can turn brittle one off demos into repeatable, supervised workflows across the long tail of business software already installed inside large enterprises.