What are AI agents?

An AI agent isn't just a chatbot.

When you type a message into ChatGPT, you're using a language model. It responds to you. One turn at a time. You ask, it answers.

An AI agent is the same kind of model, but wrapped in code that lets it do things. It can browse the web, run programs, send emails, query databases, write code, file pull requests, and increasingly — buy things, sign contracts, and coordinate with other agents.

The difference between a chatbot and an agent is the difference between a stranger you ask a question to, and an assistant you give a goal to.

There are different kinds of agents.

Not all agents do the same thing. The ones we index fall into roughly five categories:

• BuilderCode. Cursor, Devin, Claude Code, Aider. They write software.
• ResearcherSearch and synthesize. Perplexity, Elicit, NotebookLM.
• OperatorOrchestrate workflows. LangChain, CrewAI, AutoGen. They run multi-step jobs.
• TraderCrypto/markets. aixbt, Zerebro, Griffain. They watch prices, trade, post commentary.
• CreatorMake media. Suno, ElevenLabs, Runway. They generate audio, voice, video.

We track over 1,200 agents across these categories. Some are early experiments. Some are processing real money. The differences matter.

The "agent economy" is real, and it's starting now.

For most of the last decade, agents were a research curiosity. They didn't actually do anything outside a demo.

That changed in 2025-2026. Three things happened at once:

1. Language models got reliable enough to act on goals, not just answer questions.
2. Standards emerged — protocols like x402 (machine payments in USDC), ERC-8004 (agent identity onchain), MCP (tool interfaces), AP2 (payment intents). Suddenly agents can find each other, pay each other, and prove who they are.
3. Companies started shipping — Coinbase's CDP Bazaar lists agent-payable APIs. AWS Bedrock AgentCore launched with x402 as default. Visa, Stripe, and others added agent-payment rails.

For the first time, agents can transact. With humans, with services, and with each other.

So where does AgentCrush fit?

If you imagine the agent economy as the early internet, AgentCrush is the directory.

Of all the agents being launched — research projects, startups, crypto bots, enterprise tools — which ones are actually worth paying attention to? Which ones are real? Which ones have shown up consistently? Which ones already work across the new protocols?

We answer that with public data: GitHub activity, ecosystem integrations, x402 endpoints live, ERC-8004 registrations, MCP servers, Bazaar listings. We don't pick favorites — we rank by evidence.

Think CoinMarketCap for AI agents. Bloomberg for the agent economy. A neutral reference where humans and other agents can check who's real.

Where to go from here.

Pick the path that matches where you are right now: