Chapter 01
Windows 98, age four
I was three or four when a Windows 98 PC showed up at home. A buzzing CRT, a beige tower, a slow boot. I clicked everything, and when it didn't obey, I wanted to know why.
I'm still poking at machines. Better tools now.
Chapter 02
Cracks, forums and ports
Games were expensive in Argentina, so I hunted cracks for Mu Online, WoW, Counter Strike and Age of Empires. Half failed. The forums explained why, and then I had better questions: why does this game need an open port? What is my firewall blocking?
I came for the games and stayed for the plumbing.
Chapter 03
Sixteen, and ' OR 1=1
At sixteen I found ’ OR 1=1 -- on a forum and pasted it into a Mu Online private server login. It worked. I changed it myself to grab items I wanted, not my proudest moment, and kept digging until I understood the SQL behind it.
Today I write the parameterized query that would have stopped me.
Chapter 04
UTN
After that, breaking things wasn't enough. I wanted to build them. I started Software Engineering, then switched to the shorter Técnico Programador Universitario at UTN FRT so I could start working sooner. I graduated in 2021.
Tricks I'd been using for years finally had names.
Chapter 05
The 80% endpoint (Lytx)
At Lytx (video telematics, through GlobalLogic), clients pulled data one record per call. I designed a bulk endpoint: one batch request, handed to a new SQL Server stored procedure that processes the whole set inside the database. Client processing time dropped 80%.
Most performance wins are about the shape of the API.
Chapter 06
From breaking logins to selling access (Coalfire)
Coalfire is a cybersecurity company. I designed the access role, with its permissions and restrictions, that let them sell through AWS Marketplace, and moved critical endpoints from a monolith to a .NET 9 microservice that responds 30% faster under load. The kid from chapter 03 now builds the locks.
I know how logins break, so I build them carefully.
Chapter 07
A team of agents, one human
At Coalfire, in May 2026, I designed a five-agent pipeline. Transcript agent: meeting to specs. Backlog agent: specs to tasks. Developer agent: opens the PR. QA agent: tests it with Playwright. Tech-lead agent: reviews it. Any failed check sends the work back.
Agents do the work. I keep the calls you can't undo.
A one-month project on a team of three. Our project manager presented the proposal to the client, a co-worker tested the agents against the method, and I owned the agent design and the methodology. I drafted the method with Claude Sonnet and ran the proofs of concept in GitHub Copilot (GPT-4o).
That method was my first step, and I open-sourced it as openspec-tdd-schema. Today I use a much fuller version of the idea: ArggonManager, in chapter 08.
Chapter 08
Now: ArggonManager
My last project at GlobalLogic wrapped in August 2026. Since then I've been building ArggonManager, my own methodology for AI agents. Specs and acceptance criteria live in the repo, agents claim, build, review and verify the work, and there's a done gate they can't skip. Built with it: dotnet-batchyard, C# microservices.
The irreversible calls stay human.
my methodology for AI agents · TypeScript
ArggonManager
A work methodology for AI agents that lives in git. Four agent roles: delivery lead, standards reviewer, maker and verifier. Work is sized by risk, from a plain task up to spec, exploration and ADR. Reviews leave evidence on the item, and nothing closes with an unticked acceptance criterion. Waivers, claim takeovers, reopening and releases are human-only, and the tool enforces it. Runs on OpenCode, ZCode and Claude Code. 27 ADRs. This site is tracked with it.
showcase · C# · .NET 10
dotnet-batchyard
An open-source .NET 10 backend showcase: batch ingest, JWT plans and roles, CQRS payments with an outbox and RabbitMQ, and a strangler migration from a monolith to a service behind YARP, traced with OpenTelemetry. In k6 tests on a cloud VM at 500 items per request, median latency fell 93% (about 300 ms to 20 ms) and throughput rose from about 7.7k to 95k items/s.
C# · .NET 10 · SQL Server · Redis · RabbitMQ · YARP · OpenTelemetry · k6 · Docker
The labs below were built by my agents, under my specs and review.
lab · Rust
racha
A habit streak CLI, my first real Rust project. If there's no D-Bus for notifications it exits with code 2, so the systemd timer fails loudly instead of quietly.
lab · Python
guardian
A backup tool. It checks every copy with SHA-256 and keeps snapshots that never get modified.
lab · TypeScript
cuentas-claras
Splits shared expenses between friends and settles them with as few transfers as possible.
labs · Go
estanteria & suizo
Two small Go tools: one tracks the books I read, the other runs Swiss-system tournaments (still early).
Córdoba, Argentina · arggondev@gmail.com · linkedin.com/in/arggon · github.com/Arggon