CLI Needs to Happen to Optimize Agentic AI
Recently I wrote that if agentic AI is to truly gain traction, business users have to be the ones to build the agents that automate their tasks and workflows, and for larger processes to be automated, systems integration needs to take place so that disparate data stores become accessible by agents.
There is another wrinkle to driving true agentic AI and that is the Command Line Interface (“CLI”). Many of our open source tools are written as CLI tools. We just built and deployed a CLI wrapper for CEORater on Friday. Kilby was written both as a CLI platform as well as having a Web UI.
Why did we build these offerings as CLI tools, products, and platforms? The answer is simple. AI agents prefer to consume CLI tools, products, and platforms over Web interfaces. CLI is a system of bytes in and bytes out at its fundamental level, simple machine-to-machine communication. CLI is a text-based control layer, which is easier for machines to process versus GUIs. CLI:
Receive bytes from Standard Input
Transform bytes using a specific logic
Emit bytes to Standard Output
If you think of AI agents as people, the applications, tools etc. that the agents engage with should be written as CLI.
The issue with the Software industry rewriting its offerings as CLI offerings is that many Software companies recently rewrote their Software applications to embed LLMs. I am not sure that many of those companies will have an appetite to rewrite applications again so that they may readily be consumed by agents, especially with the 10YR at 5%. That said, they should. I believe agentic AI is a technology transformation where it will pay to be early.
On another note, we started to test equity trading on Kilby. Probably plain vanilla equities at first, maybe options, doubtful crypto unless people ask me to do it. It will be Alpaca’s infrastructure. We will provide the natural language interface inside Kilby’s chat, which may be prompted by text or voice. Kilby’s UI is most similar to OpenAI’s GPT app. This would be available as a CLI options as well. If you haven’t seen Kilby’s CLI, it looks like claude code (our CLI not pictured here). Given that Kilby enables investors to conduct deep research, to build sophisticated analysis, to synthesize disparate data, it only seemed logical that trade execution would be the natural next capability or “skill” to add to Kilby, which is rapidly becoming a truly unique AI harness for investors.
