I am Fabio Akita: a programmer for more than 30 years, entrepreneur, speaker, writer and former YouTuber. My LinkedIn has the long version.
I maintain more than 40 open source repositories with AI agents doing most of the typing. These are the tools I built to keep that manageable, the workflow that ties them together, and the writing where I document what I learn.
The second monitor tells me what needs attention. The first one runs one agent per project. A shared memory sits under all of them, and a gauge in the top bar tells me when to change providers.
Each tmux window is a project. I start the agent with ai-memory run and give it one line: run pr-audit and iss-audit, then run github-resolution. The skills do the rest.
Every project keeps its own memory of decisions and traps. A new session starts where the last one stopped, whichever agent runs it.
ai-usagebar shows how much of each plan's 5-hour and weekly quota I have used, and when it resets. When one runs dry, I move the work to another agent.
The skills do the heavy lifting. I read their reports and make the calls.
Ship first, fix later. AI made fixing faster and cheaper.
The memory lives outside the agent, so moving a task from Claude Code to Codex or OpenCode costs one command. The new agent reads the same wiki the previous one wrote.
In order of how much I depend on them. All three are written in Rust, MIT licensed, and install with one command.
A top-bar widget, a terminal UI and native tray apps that show quota, balance and reset time for Claude, Codex, GitHub Copilot, Z.AI, OpenRouter and 19 more providers.
yay -S ai-usagebar-bin# or, on any Linux or macOScargo binstall ai-usagebar
A Rust CLI that fetches every repository you track at once and prints what is waiting. Your own items drop to the bottom, since they wait on other people. Forks show your pull requests on the upstream project instead.
brew tap akitaonrails/tap && brew install ghpendingghpending add
My maintained fork of clock-tui: a terminal clock, timer, stopwatch and countdown. Widgets run any command and show its output under the clock, so the clock becomes a dashboard.
yay -S clock-tui-bintclock --theme nervTwo of my tools grew large enough for their own sites. The third is the set of prompts that drives the whole pipeline.
Long-term memory for coding agents: a git-backed wiki of plain Markdown that captures sessions, consolidates decisions and hands work from one agent to another.
Visit the ai-memory siteAn OS sandbox for AI coding agents on Linux and macOS. The project stays writable. Your keys, tokens and home directory stay out of reach.
Visit aijail.ioAbout two dozen agent skills, from pr-audit to release. Published so you can read the reasoning and write your own.
See how I use themEverything I try with AI ends up on my blog, with the numbers, the bills and the failures. That is also how I decide which model to use for what.
Your excitement about AI is inversely proportional to your knowledge of AI.
Sua empolgação com IA é inversamente proporcional ao seu conhecimento sobre IA.
One Rails app, fourteen sabotages based on real CVEs, and 44 models ranked by whether they notice and fix what is broken. Every major new model goes through the same table.
Skills are prompts in text files. Mine closed over a thousand PRs and issues. Read them, then write your own.
With frontier models making fewer mistakes than most developers, review bureaucracy is the last excuse to hold a deploy. My numbers since February.
A desk clock with widgets, manga readers, email, typing practice, karaoke: small tools born from small annoyances.
Five months in YOLO mode without an accident, and still no trust: BTRFS snapshots, restic backups, a sandbox and discipline.
The open OKF format, local embeddings by default, and teams working in parallel, compared with the alternatives.
When an agent is the main reader: small code, greppable names, provenance, headless tests and explicit rules.
The M.Akita Chronicles is a weekly tech newsletter and podcast in Portuguese, curated by me and commented on by M.Arvin, a very depressed robot. The podcast now also has some of my favorite blog posts as full episodes.
How it is madeFlow, Inteligência Ltda, Fernando Ulrich and a Tropical on Rails talk: hours about AI, programming jobs and the hype, plus the short cuts if you are in a hurry.
Watch themEach place gets a different side of me. Pick the one that matches what you want to hear about.
Where I am the most unfiltered. Hot takes, bait, arguments and the first reaction to whatever happened today. The account is protected, so request to follow.
The more formal side. Mostly my new blog posts and the industry news worth your time.
Entertainment only: games, retro consoles, device reviews and the fun stuff. No AI arguments here.
I no longer publish videos, but the archive holds up, and every new blog post gets shared in the community tab, which is surprisingly active.
Where everything is written down in full, in English and Portuguese. If you only follow one thing, follow this.
Every tool on this site, and a few dozen more side projects.
Write to boss@akitaonrails.com.
The tools, the skills and the benchmark are public repositories. Fork them and build the version that fits the way you work.