An AI assistant helps with the next thought. An AI agent helps carry the next piece of work.

An assistant makes the next move easier to see.

Asking an AI to clarify a strategy, improve an email, summarise a document, or generate options is useful precisely because the person remains in the conversation. They can steer the output, correct an assumption, and choose what happens next.

That direct loop is the right interface when the main work is thinking, writing, or deciding.

An agent needs a definition of done.

An agent works best when the request names the outcome, the relevant sources, the constraints, and the stopping point. “Prepare a meeting brief from these records and leave all changes for review” is a better assignment than “help with this account.”

The defined finish line lets the user inspect whether the work was useful, incomplete, or out of scope.

Live context makes agents practical in operations.

Operational work is spread across pages, accounts, documents, and internal systems.

Recreating all of that context in a conversation is slow and often strips out the filters, history, and current state that make the task understandable.

Strawberry can work with the browser workspace and connected context authorised for an assignment, then prepare work where the team needs to review it.

A repeatable agent should retain the method, not the guesswork.

When a task has been tested, save the agreed sources, checks, output shape, and approval boundary as a skill. A routine can then prepare that same type of work on schedule.

This preserves a method that the team can improve instead of fossilising an accidental prompt from one successful run.

Frequently asked questions