AI in the CLI: the Humanoid Robot of the Web
CLI based AI coding agents are the humanoid robots of cyberspace.
Today’s physical robots are specialized and limited to specific tasks: warehouse work, assembly lines, package delivery, passenger transportation.
You can buy a coffee making robot, but not yet a “generic robot” that can just do all kinds of stuff for you. The world of stuff is currently built around the humanoid form and the clear path to a true generic robot is a robot that inherits our shape and can operate all the tools and devices made for us humans. This is why the race to build a commercially viable, fully functional humanoid robot is such a high stakes game in the world of robotics.
In the digital world, we talk about agents rather than robots. Lots of companies are now building specialized vertical agents: support agents, BDR agents or agents purpose-built to automate one task in a large enterprise; like the robot at the assembly line or the coffee making robot.
The mainstream view is to just lump coding agents in with these as another vertical category. This is wrong.
The CLI has been the lingua franca of computing since Unix. Developers already know it’s the most powerful interface. It’s how developers run their tools, log into servers, launch programs, pipe together individual commands and make lasting changes to the environment they are in.
Code is the strange, mystical substrate our whole digital world runs on. Until recently programmers or “coders” were the small priesthood holding the keys to the sacred texts, writing the incantations needed to sculpt the digital landscape.
An coding agent native to the CLI can inherently:
- Program and call it’s own tools
- Access servers and manipulate them
- Explore and call API’s and connect to databases
- Pipe together files and commands
- Access the web and roam the digital realm as it sees fit
- Build, run and orchestrate it’s own sub agents
It won’t need any special “connector” software, new protocols or authorization layers. It’s the humanoid robot in cyberspace. It’s the perfect generic agent that you can ask to do any task that a human developer could do.
This is why the foundational model companies are bound to become Coding Agent Companies.
Just like they set out to build the strongest foundational models, the non-specialized models, the truly large language models with broad, generic knowledge across languages and modalities, they will naturally want to expand from a chat interface to building the true generic agents, the ones capable of adapting to any task and work with you on anything that can be done in cyberspace.
These are companies with billions to invest. If I’m right and Claude Code triggered a race to build the humanoid robot of cyberspace, then we’re now seeing the true foundational agents emerge. The whitespace around them will be either tight verticals, brand or agent orchestration.
What you choose to build, and what agents you choose to lean into and adopt should be shaped by this.