DEV Community: Leandro

You can use lteter tiiartnosopsn also kownn as tygmioaypcle, to make the lives of AI agents hrad.

Leandro — Thu, 07 May 2026 23:06:35 +0000

One of the selling points of open source software has always been that you could look at the code and make sure it does what is says it does. But then who could be bothered to read the source code to verify it was legit. Agents can do it for you now.

Leandro — Wed, 06 May 2026 22:29:32 +0000

Ambler TS: a minimal state-machine builder driven by agent skills

Leandro — Mon, 04 May 2026 04:41:31 +0000

Simply install the relevant skills into an empty folder, as explained in the GitHub repo, and describe you program (walk) to your coding agent, detailing its steps (nodes) and how they connect with each other (edges). Ambler TS' skills will help the agent write specs, code and tests for your project in a predictable and modular manner. You can easily run it using Deno.

I have been testing the library using local LLMs within the Pi Coding Agent. It is great for replacing automation scripts and playing with agentic workflow ideas.

Convention over configuration reaches a new level with agent skills. I built a tiny workflow orchestrator (ambler-ts, 52 lines of TypeScript with docs). But it is the skills that enforce the patterns used to write simple, readable and maintainable code.

Leandro — Thu, 30 Apr 2026 22:17:23 +0000

I have been experimenting with agent skills for software design patterns and the results have been very positive so far. I put together an Android (arch26) and Deno (ambler.ts) projects and the results are predictable while tests are written by default.

Leandro — Fri, 17 Apr 2026 16:12:33 +0000

Ambler: rules of engagement with coding agents

Leandro — Thu, 31 Jul 2025 21:58:22 +0000

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.

English will be to LLMs what the QWERTY layout is to keyboards.

Leandro — Fri, 25 Jul 2025 17:56:32 +0000

Should we all learn the Lojban language? It was created to eliminate logical ambiguity , which sounds great for LLMs, and also makes use of a subset of the English alphabet which could be great for keyboard ergonomics.

Leandro — Thu, 24 Jul 2025 01:55:29 +0000

Is Python to AI what Javascript is to the Web?

Leandro — Fri, 11 Jul 2025 00:00:18 +0000

I have been experimenting with Gemini CLI and successfully used it to build a small apps and automation scripts. It seems that, unless otherwise requested, Gemini tends to prefer to use Python as the programming language to use. I can see why:

Relatively simple
Multi-platform
Has a vast ecosystem of libraries
Interpreted

And surely I am missing other advantages that someone with more knowledge about it can point out. It makes sense AI will work better with some languages than others and I even expect new languages to emerge that make it easier for LLMs to work with them. However, I seem to already be building up a collection of useful code written in Python, so I may be inclined to continue to use it so I don't have to install too many different toolchains and I can reuse it when possible.

I would be curious to know if other LLMs prefer other languages, like Javascript or Go, and what experiences other developers have had.

Loving Gemini CLI for writing ad-hoc Bash and Python scripts.

Leandro — Thu, 10 Jul 2025 21:06:47 +0000

Gemini CLI and I created our first project together.

Leandro — Mon, 30 Jun 2025 20:49:28 +0000

Gemeni CLI and I created our first piece of working software together: https://github.com/argenkiwi/audini. It was a rewrite of and old project (https://github.com/argenkiwi/extereo): a Chrome Extension to make playlist from audio in the sites you visit.

Leandro — Mon, 30 Jun 2025 20:45:22 +0000