In June 2025 Gartner looked at the thousands of vendors claiming to sell agentic AI. It concluded that only about 130 of them were offering the real thing. The rest were relabelling chatbots and assistants. Gartner called this “agent washing”.
So what is it? Agentic AI is software that uses a language model to plan and carry out a multi-step task on its own. It decides what to do next, uses tools and systems to do it, and keeps going until the goal is met or a person takes over.
Chatbot, assistant, agent
Think of a loading dock. A chatbot is the clerk at the window: you ask, it answers. An assistant also fills in the form for you to sign. An agent takes the manifest, picks the boxes, books the truck and tells you when it’s done.
Three things set an agent apart.
It plans. It breaks a goal like “resolve this refund request” into steps nobody spelled out.
It acts through tools. It calls APIs, queries databases, sends email and writes to systems of record. That’s why model makers now tune for it. Google did this with Gemini 3.5 Flash in May, trading some long-context recall for more reliable tool use.
It loops. It watches the result of each action and adjusts. That makes agents useful, and also risky. An agent that misreads a result keeps acting on the misreading.
A product that does none of the three is a chatbot with a 2026 price.
| Product | What it adds | What it lacks |
|---|---|---|
| Chatbot | Answers from knowledge | No actions, no memory of goals |
| Assistant | Drafts and suggests actions | Human carries out every step |
| Agent | Plans, uses tools, loops on results | Needs guardrails and review gates |
- What it adds
- Answers from knowledge
- What it lacks
- No actions, no memory of goals
- What it adds
- Drafts and suggests actions
- What it lacks
- Human carries out every step
- What it adds
- Plans, uses tools, loops on results
- What it lacks
- Needs guardrails and review gates
Cite or embed this
How much is really in production?
Gartner’s 2026 survey of technology executives found 17 percent had deployed agents, with 42 percent expecting to within a year.
The same firm forecasts spending on agentic software will grow 141 percent in 2026 to roughly $200 billion. That’s inside a total AI market of $2.59 trillion.
The agentic figure shifts between editions: $201.9 billion in one, $206.5 billion in another. But the shape holds. Spending is more than doubling in a market where few buyers have scaled anything.
The survey funnel, and why pilots stall, are in the agent ROI piece.
Where do agents work?
Agents pay their way in three kinds of work:
Each has an outcome you can check and a small blast radius. Gartner’s own advice is to pursue agents only where the value is clear and to rethink workflows rather than bolt agents onto old systems.
The failures are open-ended tasks across many systems, with no ground truth and no review. Think booking complex business travel across six tools, or running a sales territory alone.
Gartner expects 33 percent of enterprise software to include agentic AI by 2028, up from under 1 percent in 2024. That’s growth on a timeline of years.
So ask any vendor to show you the loop. What does the agent see after it acts? What does it do when that says it failed? Where does a person come in? A real agent has a specific answer. A relabelled chatbot has a slide.
FAQ
What is agentic AI in simple terms? It’s software that chases a goal on its own. It plans steps, uses tools like APIs and databases to act, and adjusts based on results. A chatbot only answers questions. An assistant suggests actions a human must carry out. A practical test: can the system finish a multi-step task end to end without a person driving each step?
How is an AI agent different from a chatbot? A chatbot finds or writes answers in a conversation and takes no actions. An AI agent plans a sequence of steps, carries them out through tools and integrations, and reacts to the results until the task is done or handed up. Many products sold as agents in 2026 are chatbots with new branding. Gartner calls this agent washing.
Are AI agents actually being used in production in 2026? Yes, but far less than the marketing suggests. McKinsey’s survey data shows 62 percent of organisations experimenting with agents, while 23 percent are scaling one anywhere in the business. Gartner predicts over 40 percent of agentic AI projects will be cancelled by the end of 2027. Today’s production wins cluster in bounded, high-volume tasks like support triage and coding workflows with human review.
Three-minute lesson on this Workflow or agent? Marketing course, advanced. Free, no sign-up.