Generative AI creates an output. Agentic AI works through the steps needed to reach an outcome.

Generation is the first useful layer.

A model that can explain a complex concept, write a first draft, summarise a report, or produce code has already removed real cognitive work. For a self-contained task, generation may be exactly what the user needs.

The limitation appears when the next step is outside the answer: inspect the latest account, compare a set of documents, prepare the CRM update, or assemble the handoff for someone else.

Agency begins with context and a finish line.

An agent needs more than permission to call tools.

It needs a bounded objective, the relevant sources, a definition of done, and instructions for uncertainty.

For example, “prepare a renewal risk brief from these tabs, cite each claim, and do not contact the customer” gives the system a goal and a stopping rule. The result can be assessed against the request rather than the fluency of its prose.

Tools make the output operational.

Generative capability can be part of an agentic workflow: it may write the draft, explain the exception, or shape the report. Tool use lets the workflow gather current information, organise material, and prepare work in the place where a team needs it.

Strawberry is built around browser context, open tabs, files, signed-in sites, and connected apps, which makes it useful when an assignment starts from live work rather than a self-contained prompt.

Autonomy has to stop before accountability disappears.

The value of an agent is not that it acts without people.

It is that it reduces the work between a request and a decision-ready result.

For actions with consequences, such as sending a message, changing a record, publishing content, making a payment, or deleting data, the workflow should surface the final choice to an accountable person.

Frequently asked questions