One thing I love about AI agentic coding assistants is how they let me try and fail so much quicker and safer without frustrating me.
Now I can freely, safely, and easily start working on ideas, weird directions, even complex directions, from a new project to choices during algorithm development. Love it!
Just found out Anthropic acquired Bun — interesting.
Let’s see how fast Claude Code will be super super soon. :3
https://www.anthropic.com/news/anthropic-acquires-bun-as-claude-code-reaches-usd1b-milestone
Good news: the GitHub Action — and yes, it’s approved by my company — now supports my flows. There are still hiccups in the runtime and the pre-indexed strategy, but they’re tiny. The biggest elephant in the room is solved, can’t wait to bring all my ideas on.
I’ve been talking about the VM-operated AI agents idea, but my company policy is very strict about VM networking, which is preventing me from building the helper agents I wanted.
I’m in the product software engineering team, working on this as a side project for my team — not in the DevX team that gets exceptions anyway. So today I’m switching to using GitHub Actions to run the AI agents in my free time instead. Wish me luck!
I’m using CodeRabbit and Cursor Agent Review to look over my work before submitting it. They’re great, but it still feels like it’s not enough for me.
I’m thinking about re-reading books like Refactoring and that pragmatic software engineer book to write myself a guideline for how I want my work to be reviewed.
The idea is to let AI agents use that guideline to review my work for me. I’m imagining multiple agents spinning up at the same time, each focusing on a different space: correctness, conventions, performance, security, etc.
With the power of Claude Code Max x20, I just hope the quota is large enough for all of this 😭
I’m thinking about a NotebookLM look-a-like, but with confidentiality in mind.
The idea is to have a VM that hosts Codex CLI (which I’m allowed to use at work). Each “project” is just a folder that I’ll prepare manually at first (and later automate) with all the project-related documents, repositories, and other context.
On top of that, I could build a simple chat bot or StackOverflow-like interface that triggers Codex CLI to jump into the right folder and answer questions using only what’s inside that folder.
This way, I still get rich, context-aware answers without depending on non-approved tools — and it’s very cost-effective because it’s all powered by a single agents CLI tool.
I recently learned about SCIM while working on a project at Axon.
Lesson learned for my own personal products: just use WorkOS or something similar — SCIM is breaking complicated 🤯
Besides having a script to init a new repository for my service like before, I’m testing creating a new service using Cursor’s rules and guidelines.
First impression: very good — and it cost less than $0.30 to get a ready-to-code repository, thanks to the AI agents.
Today I set up a small n8n workflow that lets me turn a Telegram message into a blog post — including this one.
The flow is simple: I drop a rough idea into Telegram, n8n picks it up, sends the text to Codex CLI, and lets it shape everything into a short, readable post before committing it to this repo and pushing it live.
What I love most is how little friction there is now. I don’t have to open an editor, think about formatting, or worry about polishing every sentence. I just write like I’m texting a friend, and the automation handles the rest, which makes it way more likely I’ll actually publish quick updates like this.
I’ll write a proper article soon that walks through the whole setup — from the Telegram trigger, to how Codex CLI fits into the pipeline, to how the final Markdown post ends up on the blog automatically.
Here we go — I’m starting a new blog.
The idea is simple: a place to share my journey as a software engineer, and also just as a person navigating life. Not everything has to be about code, right?
I’m hoping this becomes a useful notebook for myself. A way to document projects, ideas, and the occasional struggle. And hey, it might even come in handy during interviews when someone asks about my past work — instead of stumbling through memories, I can just point them here.
Let’s see if I can keep this going for a while. Wish me luck! 🤞