Ask ten people what an “AI agent” is and you’ll get ten answers. That’s a problem — because the word is now stretched over everything from a chatbot with a new coat of paint to software that can run a business process end to end.
Gartner has a name for the gap: “agent washing.” A lot of what’s being sold as an “agent” today is really an assistant with better marketing. And the interest is real — Gartner’s Hype Cycle for Agentic AI notes that far more organisations plan to deploy agents in the next two years than have actually done so. So it’s worth being precise about what we mean, because the distinction changes how you think about work itself.
The difference isn’t the technology. It’s who’s holding the pen.
Assistants and agents share the same plumbing — they understand language, work with information, make suggestions. From the outside they can look identical. The real difference is responsibility. An assistant hands you information and options; you still decide and act. An agent can decide what needs to happen next and do it — completing steps inside a real process, and pulling a human in when approval matters.
What that looks like when it’s real
I’ll skip the textbook example and give you one I’ve actually built.
GP surgery receives referrals by email, in no fixed format, into one shared inbox. Caseworker reads each one, judge how urgent it is, and re-type it into an internal case system 15 minutes each, with a backlog that can run to days. Some of the most skilled people in the building spend their day copying and pasting.
An assistant could help them write faster or summarise a policy. An agent sits inside the process. It reads the unstructured referral, pulls out the fields, sets a provisional urgency, and hands it to a caseworker to approve. On the nod, it enters the record into the legacy system automatically. The caseworker still owns the clinical judgement. The agent owns the donkey work.
That’s the shift. It isn’t AI writing a nicer summary. It’s AI completing a business process, with a human where a human belongs.
Why the foundations matter
The moment software starts acting on your behalf, governance stops being a nice-to-have. Who approved what? What can the agent touch? Where’s the audit trail? Get that right and autonomy becomes something you can actually trust.
Continue the conversation
That’s the theme of our webinar, Autonomous AI Agents: Building the Foundations for Success, on Wednesday 9 September, 10–11am. My colleagues and I will show what agents look like in practice, how to govern them, and what “ready” actually means — including a live look at the example above.