Browserbase explains an AI agentAn AI system that doesn't just answer once but works toward a goal in a loop — taking actions, reading the results, and deciding what to do next.Full definition →-design choice through what the agent can perceive. Screenshot-based computer use locates controls visually and acts through mouse coordinates or keystrokes. Structured browser use reads the page's representation of elements, such as buttons and fields, and targets those elements directly. Stable steps can then run as ordinary code, reserving model decisions for steps needing interpretation. The comparison recommends choosing the interface around the task: page-based interaction for web workflows, desktop control when the work requires native applications or operating-system controls.
Key takeaways · AI-distilled
Browserbase draws the line by perception: computer-use agents read screenshots and act at pixel coordinates, while browser-use agents read the DOM and accessibility tree and act on element handles that resolve the same way across runs.
The cost argument is that computer use pays a vision inferenceRunning a trained model to get answers — the phase where AI is actually used, as opposed to trained.Full definition → on every step, whereas a structure-driven agent can run deterministic steps such as navigation or clicking a stable button as plain code and call a model only where the page needs judgment.
Browserbase recommends computer use only when a task leaves the browser: native desktop apps, installers, file dialogs, OS settings, or software with no web surface or API.
The article's '~99% of agent work is web work' figure is explicitly illustrative, not measured, and the piece is written by a company that sells cloud browsers and the Stagehand SDK.
Terms in this piece · Glossary
AI agent — An AI system that doesn't just answer once but works toward a goal in a loop — taking actions, reading the results, and deciding what to do next.
inference — Running a trained model to get answers — the phase where AI is actually used, as opposed to trained.
