The 130 Vendors Behind "Agentic AI"

Thousands of vendors are selling you "agentic AI" right now. Gartner looked at the actual market and found that roughly 130 of them have real agentic capability. The rest took an existing chatbot, an old RPA tool, or a plain AI assistant, and relabeled it. Gartner has a name for this: agent washing. I'd call it something simpler. Most of the industry is selling you a new coat of paint on a tool you might already own.

I don't think you need to write a line of code to catch this. You need one plain-language question and the nerve to ask it in the room, in front of the vendor's sales team.

Here's the actual difference between a real agent and a repainted chatbot. A real agent runs a loop. It plans a step. It acts through some tool or system. It observes what happened, and adjusts its next move. It carries memory of what it already tried, so it doesn't repeat the same failed move twice. A scripted bot, or an old RPA tool with an "agent" sticker on it, can't do that. It follows the same fixed path every time. When that path hits something unexpected, it stalls, loops, or just hands the task back to a person.

So ask for the demo that actually tests this. Not a scripted walkthrough on the vendor's own sample data. Your data, live, with a deliberate failure built into the first attempt. Watch what the tool does next. A real agent notices the failure and tries something different on its own. Everything else either freezes or falls back to a human without saying so. That's fine, as long as the vendor admits that's what's happening instead of calling it autonomy.

Memory matters more than most executives assume going in. Without it, an agent hits the same wall every time it shows up. It has no record of what it already tried and failed. Ask the vendor directly what the agent remembers between one step and the next. Ask whether your team can actually see that memory, or whether it's locked inside the vendor's own system where you'll never get to audit it.

Gartner's other number here should worry you more than the vendor pitch does. The firm projects that over 40% of agentic AI projects will be canceled by the end of 2027. Cost, unclear value, and weak risk controls are the stated reasons, not failed technology. Buy a repainted chatbot believing it's a real agent, and you've bought a head start on exactly that cancellation. Nobody on your team will figure out why for months.

Deloitte's research adds the piece that makes this urgent, not academic. Only 21% of enterprises have a mature governance model in place for agentic AI today. Yet 74% expect to be using AI agents at least moderately by 2027. Nearly a third of those expect extensive or full integration. Most companies are racing to expand something they haven't yet learned to govern. A market full of relabeled tools only makes that governance problem harder to see coming.

I think about this the way I think about every wave of enterprise technology that got oversold before anyone could verify the claims. The dot-com years taught me the same lesson from the inside, at Napster: capital moves fast, and a label moves faster than capital. "Agent" is becoming one of those labels. It gets stamped on anything with a chat window and a task list, whether or not anything underneath actually plans, acts, and adjusts on its own.

The human-plus-agents stack only works when people stay in the loop on purpose, not by accident because the tool couldn't handle the moment it hit. Your team needs a plain, direct answer. Can what you bought actually adjust when things go wrong, or does it just hand the task to a person and call that graceful design?

Before your next agentic AI purchase, ask one question, and don't accept a vague answer. What happens when the agent is wrong? If the vendor can't show you that moment, live, on your own data, you already have your answer about which side of the 130 they're really on.

Sources: Gartner, June 2025 (agent washing, 130-vendor estimate, 40% project cancellation forecast) · Deloitte, State of AI in the Enterprise, January 2026 (governance maturity and deployment plans).

Alex translates agent architecture into the questions a board actually needs to ask before it buys.

book a conversation →