Geoffrey Huntley argues that well-designed command-line interfaces (CLI) are sufficient for AI agents to perform tasks without needing additional protocols like MCP. He classifies developer tool companies into two tiers based on their presence in AI model weights.
Companies in the S-tier are already integrated into model weights, allowing agents to operate using standard CLI tools. GitHub is cited as an example where models can execute complex repository management tasks directly via its CLI.
For companies not yet in model weights, Huntley suggests optimizing the --help documentation of their CLI. He recommends refining commands to ensure agents can reach results with minimal tool calls, potentially using automated testing cycles.
The final stages involve optimizing documentation for agentic search engines (ASEO) and pursuing commercial agreements with AI labs to include documentation in future model training runs, bypassing the need for users to install extra skill packages.
