
Gartner says only about 130 'AI agent' companies are real
Gartner Says Only About 130 “AI Agent” Companies Are Real. I Build Agents. Here’s How to Tell.
以下正文同步自 Hacker News 最新,版权归原站所有,已转换为易读排版。
Artificial Intelligence
Startup
Technology
Entrepreneurship
Gartner Says Only About 130 “AI Agent” Companies Are Real. I Build Agents. Here’s How to Tell.
Agent washing is everywhere. Here’s the 5-question test I use to spot it.
Devansh Ranjan
4 min read·
Just now
--
Gartner estimates only about 130 of the thousands of agentic AI vendors are real. Illustration by the author.
Thousands of companies now sell “AI agents.” Gartner thinks about 130 of them are real.
That’s not my number. In a 2025 forecast, Gartner estimated that only about 130 of the thousands of agentic AI vendors are the real thing. The rest are doing what it calls agent washing: putting a new label on chatbots, AI assistants and old automation tools.
I’m 18 and I run Sumora, “Shopify for AI agents.” So I have an obvious stake in this. But I’d rather tell you how to spot a fake than watch my whole category turn into a joke.
The same Gartner report predicts that over 40% of agentic AI projects will be canceled by the end of 2027, because of rising costs, unclear business value or weak risk controls.
Meanwhile, adoption keeps climbing. In an IBM study published this week, 35% of 3,221 senior executives at large companies said they already run agentic AI in production. Only 37% said their governance is built to scale with it.
Companies are buying agents faster than they can tell which ones are real.
Carnegie Mellon researchers built TheAgentCompany, a fake software company where AI agents get 175 real office tasks: browsing internal tools, writing code, messaging coworkers.
The best agent they tested, running on Gemini 2.5 Pro, fully completed 30.3% of tasks. It took almost 27 steps and more than $4 per task.
The failures are the interesting part. One agent got stuck because it couldn’t close a popup. Another couldn’t find the right coworker in the chat app, so it renamed a different user to that person’s name and messaged them instead.
Figure 1: In TheAgentCompany, the best agent fully completed 30.3% of 175 tasks. Data: Xu et al., arXiv 2412.14161.
That’s what real agents look like today: useful, but messy. So when a vendor says their agent handles everything perfectly, one of two things is true. It isn’t really an agent, or nobody has tested it properly.
An honest agent company can tell you exactly where its agent breaks.
Strip away the demo and an agent is a loop, not a reply:
while goal not met: look at the current state decide the next step act with a real tool check whether it worked if stuck or risky: ask a human
A chatbot gets a prompt and returns text. An agent gets a goal and keeps going until it’s done or it needs you. These five questions show you which one you’re looking at.
If a product fails the first two questions, it’s not an agent. It’s a chatbot with a new homepage.
Our outbound agent starts from a goal: who you want to reach. It finds and researches prospects across LinkedIn and Instagram, looks for signals, ranks them, and builds a personalized lead magnet for each one, like an Instagram audit or a competitor content breakdown. Then it writes the outreach.
You stay in control of what actually gets sent. That’s on purpose. Sending a message as your business is exactly the kind of step that should go back to a human.
And it’s measured on one outcome: 3 meetings a week. Six customers pay $500 a month for it today. I wrote about how we got to that price in I Raised My Price From $99 to $500 a Month.
Agents are real. Most of what’s sold as one isn’t yet.
Before you buy an AI agent, ask what it does, who decides the next step, when it asks you and what result it’s paid for.
If you run an agency or a B2B business and want to see what a working outbound agent looks like, book a 15-minute call with me here. Bring your hardest questions. I’ll show you where ours is strong and where it isn’t.
And if this helped you see through the hype, clap so another founder finds it, and follow for the next breakdown.
正文由 探How 从来源站点 RSS 同步,内容未经改写;遇到排版缺失或需要图片、视频时请以原文为准。