AI browser automation, term by term

Software running inside a browser that can read the current page and carry out requested tasks through connected app APIs or the website interface.

It differs from a chatbot by having page context and the ability to act, and from an autonomous agent by keeping a human in the loop for consequential steps.

A pattern where an AI system plans several steps, navigates pages and adjusts based on what it observes, rather than answering in a single turn.

Useful for multi-page research and long forms; the risk it introduces is unverified completion, which is why state should be read back after each write.

A defined set of operations that call a service's API through a backend, using credentials bound to your account.

In resolvenengine each connector operation carries its own Allow, Ask or Deny policy.

A task performed through a website's own interface (clicking, typing, navigating) when no direct API operation exists.

Slower and more fragile than a connector, so it is the fallback rather than the default for your own data.

A service exposing tools over the Model Context Protocol so an AI client can discover and call them in a standard way.

Remote MCP servers can be registered as connectors, with private-host protection and per-tool permissions.

A confirmation step shown before a write, describing exactly what will change and where.

Approval is required for any operation you set to Ask, and a Deny is never bypassed.

A pricing model where the product charges for the software and you pay the AI provider directly for model usage with your own API key.

It removes the reseller margin on tokens and lets you pick a stronger model for the same money.

A request phrased about your own accounts ("my inbox", "in my workspace") which routes to a connector instead of the open web.

Naming a product alone is a weak signal and usually means research, not account access.

Checking the resulting state after an action, instead of treating a click or a keystroke as proof that the task finished.

For form filling this means reading the live fields back and retrying only the differences.

The lag between an event happening and it appearing in an analytics product's reports.

Search Console data commonly lags 2–3 days and GA4 standard reports settle over roughly 24–48 hours, so comparisons should use complete days.