Agent, Assistant, or Autocomplete? A Plain-English Field Guide

A vendor emails you about their new AI agent for scheduling. You open the demo and it looks like a chatbot with a calendar widget. The pricing is triple what you paid last year for the same tool. You are not sure whether you are looking at an actual agent, a rebrand of last year's assistant, or a slightly smarter autocomplete.

You are not the only one confused. Gartner has a name for this problem now. They call it agentwashing, and they estimate that of the thousands of vendors labeling their products as agentic, only about 130 are actually building agents. The rest have relabeled what they were already selling.

That leaves small business owners in a hard spot. You are being asked to pay more for AI-labeled tools, and there is no easy way to tell what you are actually buying. Here is a plain-English breakdown.

The Three Categories Everyone is Blurring

There are three genuinely different things being sold under the AI banner right now. Getting them straight is the first step to not overpaying.

• Autocomplete. Predicts the next word, sentence, or field. You are still doing the work. It just fills things in faster. Grammar suggestions, email subject lines, spreadsheet formulas.
• Assistant. Answers questions and drafts things when asked. It waits for you. It does not act on its own. Chat interfaces, drafting tools, Q&A over documents.
• Agent. Takes a goal, plans the steps, uses tools, makes decisions, and completes the whole workflow. You review the result, not each step. Invoice processing, appointment booking, expense categorization.

The difference is not about how smart the model is. It is about who is doing the work. In autocomplete, you are. In assistants, you and the tool trade turns. In agents, the tool does the whole loop and hands you the result.

The Four Questions That Tell You Which One You Are Buying

Ignore the marketing. Ask the vendor these four questions in order. The answers will place their product in one of the three buckets above.

1. Does It Take an Action or Just Suggest One?

An agent completes the task. An assistant makes recommendations you then execute. If the product's job ends the moment it hands you a draft to copy and paste, you are looking at an assistant, not an agent.

For an invoice tool, an assistant reads the PDF and shows you the fields. An agent reads the PDF, matches it to a PO, checks for duplicates, updates your accounting system, and puts a payment recommendation in your approval queue.

2. Does It Use Tools, or Just Talk?

Agents call other software. They log into your calendar, hit your accounting API, send an SMS, query a database. Assistants mostly generate text. If the demo shows a chat window and nothing else, you are looking at an assistant with a nice interface.

Ask the vendor to name the specific tools their product connects to and the specific actions it takes in each one. If the answer is vague, the connections probably do not exist.

3. Can It Handle a Multi-Step Job Without Being Prompted at Each Step?

An agent takes "process this month's invoices" and works through 40 of them. An assistant needs you to prompt it 40 times. If the product needs you to type "next" or paste the next document, it is an assistant.

The keyword to listen for is workflow. A workflow implies steps that run in sequence without human input between them. Chat is not a workflow. A queue of drafted actions is.

4. What Does It Do When It Is Unsure?

Autocomplete guesses. Assistants ask you to clarify. Agents flag their uncertainty with a confidence score and put the uncertain item in a queue for human review while the certain items proceed.

If the vendor cannot explain what happens on ambiguous items, or worse, tells you the AI is confident every time, you are looking at a product that will silently make bad decisions. The good agents are the ones that know when to stop.

Why This Matters for What You Pay

The pricing gap between these three categories is real. Autocomplete features are increasingly bundled into software you already own for free. Assistant products typically run $20 to $50 per user per month. Agent products run $100 to $500 per month per workflow, sometimes more, and are priced against the labor cost they replace.

When a vendor rebrands their assistant as an agent and prices accordingly, you are paying the labor-replacement price for a product that does not actually replace any labor. The reason Gartner expects over 40% of agentic AI projects to be canceled by the end of 2027 is that a lot of buyers are figuring out mid-contract that they bought an assistant with an agent price tag.

What to Look For Instead

Real agents show three signs in a demo:

• The demo runs end-to-end without a human typing between steps. You watch it complete a whole invoice or a whole booking loop.
• There is a queue of completed-but-unapproved actions. This is where the agent's outputs land for you to review.
• The pricing story matches an outcome, not a seat. Priced per invoice processed, per appointment booked, per hour saved. Seat-based pricing usually points to an assistant.

If the demo is a chat interface, the pricing is per seat, and the vendor cannot show a workflow completing on its own, you are almost certainly looking at an assistant with an agent brand.

What to Do Next

1. Reread your last two AI product proposals with these four questions in hand. Highlight the answers you did not get. Send the vendor the specific list of unanswered questions and see how they respond.
2. Look at the pricing model. Per seat suggests assistant. Per outcome suggests agent. Ask the vendor which they are selling and why the pricing matches.
3. Take the AI Readiness Assessment. Four minutes, no signup. It maps your current processes to the category of AI that actually fits, so you stop overpaying for the wrong tier.

The AI category matters because the price matters. An assistant is a fine purchase when you need an assistant, and an agent is a fine purchase when you need an agent. What is not fine is paying for one and getting the other. Ask the four questions, and the marketing labels stop mattering.