How I work with AI agents
Almost everything in this portfolio was built and is operated with AI coding agents (Claude Code), across several sessions a day. Each session takes a request, drafts a step-by-step plan, executes it, verifies the result and closes with a summary of what is done and what is still open. This page shows the log of those closings, numbers only.
Tasks closed per day
Where the work went, last 30 days
What is counted
- Task closed: a request the session finished and summarized, stating what changed and what was not verified yet. Requests dropped halfway do not count.
- Commit: commits made that day in the systems’ repositories (all private, except this portfolio).
- Production deploy: releases to the VPS where the systems run, grouped per session.
- Area: each task is classified by the subject of its own summary: systems and ERPs, infrastructure and security, company administration, marketing and websites, career and studies.
What stays out
No task text, session title, customer, amount or infrastructure detail leaves my computer. The full dashboard, with the daily feed, is private. Only a file with daily counts comes here, generated by a script and published by a scheduled job.
Why publish this
Because it shows the method better than a list of tools: a human in the loop, a plan before execution, verification before closing and a record of what was done. The Genesis, Clinalta and Escrituro case studies show the outcome of this method as products.