What Claude Code really is
By Miguel Torrez, founder of Genial Agency
You will be able to tell the AI model (the brain) apart from the app around it (the body, or harness), and choose between Claude Chat, Cowork, Code, Hermes and OpenClaw based on what each body lets the model do.
What you need
No setup needed, this is a concept video
Optional: a Claude account at claude.ai to compare Chat, Cowork and Code yourself
Step by step
Spot the shared name
Look at Anthropic's three apps: Claude Chat, Claude Cowork and Claude Code. Notice that every one of them carries the word Claude. That word names the brain inside the app, not the app itself.
Treat the LLM as the brain
Think of the LLM as a brain on its own. By itself it cannot really do anything: it only gives you the next most probable word when you ask it a question.
See each app as a body
Picture Chat, Cowork and Code as three different bodies (I call them Iron Man suits) built around the same Claude brain. The app is what gives the brain something to act with.
Compare capacities, not intelligence
Remember that the brain is literally the same in all three Claude apps. What changes is what it can do, because it sits inside a more capable body. Pick the app by the capacities you need.
Place Hermes and OpenClaw
Recognize Hermes and OpenClaw as bodies too. They are not made by Anthropic but by outside developers. Links: Hermes Agent docs and OpenClaw.
Hermes Agenthttps://hermes-agent.nousresearch.com/docs/
OpenClawhttps://openclaw.ai/
Swap the brain when you need to
Use Hermes, OpenClaw and similar agents when you want to choose the brain: they can host an Anthropic, GPT or Gemini model. Anthropic's own apps are private bodies where you cannot put in any brain you want.
Learn the word harness
Call everything around the brain the harness. ChatGPT, Claude Code and Claude Cowork are all harnesses. Once you are advanced enough, you can build your own body for your use case with a platform such as Pi.
Judge apps by the body
Expect very different results from the same brain in different apps. With today's models, the body matters more than the brain: compare harnesses by the capacities they give, whether off the shelf from Anthropic or through OpenClaw, Hermes or future providers.
Watch out for
Assuming Claude Chat, Cowork and Code run different models: the brain is the same, only the body changes.
Choosing a tool by model intelligence alone: the same model gives very different performance depending on the harness.
Expecting to put a GPT or Gemini brain inside Anthropic's own apps: those are private bodies, unlike Hermes or OpenClaw.
