A chat assistant answers the message in front of it. An agent takes a goal, makes a plan, works through the steps, checks its own results and keeps going until the job is done or it needs your input.
The difference is doing versus answering. Ask a chat assistant "what should a good weekly report contain?" and it tells you. Give an agent access to your analytics and say "prepare my weekly report", and it pulls the numbers, writes the report and hands it to you for review.
To do that, an agent needs tools: the ability to browse, read files, run software, connect to your systems (often via MCP). What it is allowed to do on its own, and when it must ask permission, is decided by its harness.
Why you keep hearing it
Because the industry has spent the last couple of years shifting from chatbots to agents, and every product announcement wants a piece of the word. Some things called agents genuinely plan and execute multi-step work. Others are a chat window with a new label.
What it means for you
Ignore the label and ask what it actually does. Can it complete a real task from start to finish? Does it check its own work? What happens when it gets something wrong? The honest answers to those questions matter more than whether the word agent appears on the pricing page.