This content originally appeared on DEV Community and was authored by Leandro
I recently came across an interesting project called PocketFlow, which presents a minimalistic approach to building workflows that incorporate large language models. It is originally written in Python, a language I have never really used, so I attempted to port it to Kotlin to better understand how it works. As I progressed it became clear the project was simply a glorified state machine and it could be simplified even further. The result was Ambler: a very simple function and a very simple class definition that allow you to express a program as a series of steps that update the current state and pass it on to the next step.
Nothing groundbreaking, but the power of this simplicity is that you can describe your application logically in plain English in a markdown document and then ask a coding agent to build it for you, while keeping the application structure consistent, predictable and easy to understand.
I decided to build a simple counter application as the first example. By not specifying the programming language, Gemini CLI tended to gravitate to using Python. I went along and refined the approach until I got exactly what I wanted. By the end of it I also obtained an equivalent implementation of the Ambler code and the sample in Go, JavaScript, Kotlin, Ruby, Rust and Typescript.
I have observed something interesting when looking into the project on GitHub: the percentage of the codebase written in each language varies considerably for an equivalent implementation. It makes me wonder what the long term impact of using a less verbose, more concise programming language has on your agentic coding costs, assuming there will be a direct correlation between the amount of code needed to solve a problem and the number of tokens used.
I'll continue experimenting with this approach, but I can already see how useful it will become when needing to automate simple tasks. Gemini CLI has already made me a handy application to download URLs from an m3u file which is a great companion to an Chrome Extension Gemini CLI built for me a while ago. All I need now is more ideas to test the approach with.
This content originally appeared on DEV Community and was authored by Leandro
Leandro | Sciencx (2025-07-31T21:58:22+00:00) Ambler: rules of engagement with coding agents. Retrieved from https://www.scien.cx/2025/07/31/ambler-rules-of-engagement-with-coding-agents/
Please log in to upload a file.
There are no updates yet.
Click the Upload button above to add an update.