The AI agent landscape: builders, products, runtimes, and frameworks
"Agent" now spans no-code canvases, finished products, cloud runtimes, and developer libraries. Two questions — who it's for, and what they do with it — sort the whole confusing space into four clean categories.
Ask ten people what an 'AI agent' company does and you'll get ten different answers — because the word now stretches across no-code canvases, finished products, cloud runtimes, and developer libraries. The confusion clears the moment you stop asking 'is it an agent?' and start asking two better questions: who is it for, and what do they do with it? Sort every offering by those two axes and the whole landscape falls into four clean categories.
The two questions
The first axis is who: does getting value out of it require engineering skill, or not? The second is the verb: are you assembling an agent, using a finished one, running agents in production, or writing agent code from scratch? Cross the two and you get four categories — Builders, Products, Runtimes, and Frameworks — each with a persona that fits it.
Builders — non-technical people build agents
Builders are visual, low-code environments where someone without an engineering background can assemble a working agent by wiring nodes and configuring steps. You reach a running agent in an afternoon, no codebase required. Tools in this bucket include n8n, ElevenLabs, the OpenAI Agent Builder, and CrewAI Studio (I've written a post on each). The persona is the AI Builder: a founder, ops lead, or product person shipping automations without waiting on engineering.
Products — non-technical people use agents
One step removed from building: Products are finished, agentic tools you simply use. You're not assembling the agent — someone did that for you — you just point it at your work. Coding and knowledge-work products like Claude Code and its peers sit here: agents packaged as products. The persona is the AI Coder, someone getting real work done through an agent without ever opening its internals.
Runtimes — technical people execute agents
Now we cross into engineering. Runtimes are where an agent actually runs in production — the managed infrastructure that handles sessions, state, scaling, and isolation. AWS Lambda, Amazon Bedrock AgentCore, Anthropic's Managed Agents, and Google's Vertex AI Agent Engine live here. The persona is the AI Engineer on the production track: less concerned with which framework wrote the agent, more with running it reliably and cheaply at scale. I go deep on these in the agent-runtimes post.
Frameworks — technical people develop agents
Frameworks are the code-first libraries an engineer uses to build an agent from scratch — the loop, tool calling, orchestration, and handoffs. The OpenAI Agents SDK, CrewAI, LangGraph/LangChain, and Google's ADK are the main options. The persona is the AI Engineer on the agentic track, who cares about control, testability, and how the agent actually thinks (see the agent-framework-landscape and LangGraph posts for the detail).
Which one are you?
The point of the map isn't to rank the categories — each is right for a different person and job. It's to locate yourself:
- AI Builder — you want a working agent without writing code: start in a builder.
- AI Coder — you want an agent's output, not its internals: reach for a product.
- AI Engineer, production track — you need to run agents reliably at scale: pick a runtime.
- AI Engineer, agentic track — you're building the agent itself: choose a framework.
Most of the noise in the agent conversation comes from two people using one word for two different layers — a founder who means a no-code builder talking past an engineer who means a framework. Name the layer first, and the right tool, the right skills, and the right expectations all follow.
There isn't one 'agent' market — there are four: for people who build, who use, who run, and who develop. Progress starts with knowing which one you're standing in.