"Agent" is the most overused word in software right now. One analysis of U.S. Google search data found that searches for "agentic AI" averaged about 110,000 a month and were up 39% year over year, with "AI agents" at about 60,500 a month.[2] (That analysis is a secondary summary of the underlying data, so treat the figures as approximate.) When a word gets that popular, vendors attach it to everything. Before you pay for one, know which of four things you are looking at.
The four categories
| Type | What it does | Good for | Main risk |
|---|---|---|---|
| Rules-based automation | Does the same steps every time something happens (new form, new invoice, paid deposit) | Predictable, repeatable work | Breaks when inputs change shape |
| Chatbot | Answers questions in a chat window, usually from a fixed set of content | FAQs, simple intake, after-hours questions | Confident wrong answers; dead ends for the customer |
| AI assistant | Helps one person draft, summarize or analyze when asked | Writing, research, summarizing documents | Output quality varies; staff paste sensitive data into it |
| AI agent | Takes a goal, decides the steps and uses tools to carry them out | Multi-step work with judgment, like triaging a request and acting on it | Takes wrong actions; harder to test and audit |
Most owner problems are solved by the first row. An invoice arrives, the amount and vendor are read, the invoice is coded to a job and routed for approval. That is automation with a bit of AI for reading the document. You do not need an agent for it, and a simpler tool is easier to test and cheaper to run.
Agent washing is real
Gartner uses the term "agent washing" for the rebranding of existing products, such as assistants, robotic process automation and chatbots, without real agentic capability. It estimates that only about 130 of the thousands of vendors claiming agentic AI are the real thing.[1] The same release attributes the expected cancellations to cost, unclear business value and inadequate risk controls. Gartner's supporting poll was of 3,412 webinar attendees, which is not a random sample, so this is an analyst's read of the market, not a census.[1]
Five questions to ask in any demo
- What decides the next step? If the answer is a fixed flowchart, it is automation. That is fine, but price it as automation.
- What can it do without asking me? Get the exact list of actions: send an email, change a record, issue a refund, book a job.
- What happens when it is unsure? A good system stops and hands off to a person. A bad one guesses.
- Can I see a log of every decision it made? If something goes wrong, you need to reconstruct why.
- Will you run it on my data, live, today? Demos on clean sample data tell you little.
Match the tool to the problem
- The same steps, every time: automation. Cheapest and most reliable.
- Questions with known answers: a chatbot connected to your real documents, with an easy path to a human.
- A person needs to write, summarize or analyze faster: an assistant, plus a clear policy on what data can go in. See our piece on writing a one-page AI policy.
- A multi-step job with judgment, such as triaging inbound requests: an agent, but only with an approval gate on anything involving money or commitments.
Owners surveyed by Bluevine said their top uses of AI were data analysis and business insights (39%), marketing and sales (37%), automating operations (29%), customer service (27%) and financial management (23%).[3] Notice how few of those need a fully autonomous agent. Most are assistant or automation work.
Buy the least autonomous tool that solves the problem. Move up the list only when a simpler tool cannot do the job, and add an approval step each time you do.
Sources
- Gartner press release, June 25, 2025
- Truelogic, summary of a 12-month U.S. Google search-volume study (secondary source)
- Bluevine, "2026 Small Business AI Trends Report"
Figures are quoted from the sources above and were reviewed in October 2026. Where a source is a secondary summary of other research, we say so in the text. Survey results reflect each survey's own sample and definitions. Examples labeled as illustrations use made-up numbers to show the method. Nothing here is a forecast or a guarantee of results.
Want this worked out for your business?
Tell us what is wasting your time.