Matheus Castelliano
← Home

How I work with AI

AI drafts. I decide.

AI is in every step of my work, from the first spec to the code. Not as a shortcut, but as a team member I brief, review and hold to a standard.

What changed

  • Writing a scope assessment 2–3 days 1 day
  • Breaking an epic into features and backlog items 1 day

The loop

  1. 01

    Spec

    I write what we’re solving and how we’ll know it worked.

  2. 02

    Draft

    AI drafts the backlog item, the document or the code.

  3. 03

    Review

    I check every claim, number and line against the source.

  4. 04

    Ship

    Nothing goes out without my sign-off.

  5. 05

    Measure

    Then I look at what actually happened, and adjust.

01

At work

As a product owner, AI helps me write backlog items, scope assessments and acceptance documents, follow the sprint and keep cards up to date. It reads and writes directly in Azure DevOps and Jira, so context never has to be copied by hand.

One rule holds everywhere: AI drafts, I decide. Nothing is published to a board or sent to a client without my review.

02

My setup

Claude and Claude Code do most of the work. Obsidian is my second brain: notes, decisions and project history that become context for every conversation.

On top of that I build my own skills: reusable instructions that make the AI work the way my team works. One sketches a backlog from epic to features to items. One answers questions about the sprint in plain language. One estimates tasks with three-point estimates. One writes stories in the team’s exact format. I’m cleaning them up to share on GitHub.

03

Beyond work

AI isn’t only a work tool for me. I built a personal finance dashboard that updates itself through Open Finance, and small tools like Apolo and Artemis that I use every day.

See the tools I built →

04

What I’m exploring now

Spec-driven development: write the spec first and let it guide the AI through the code. It’s where product thinking and building meet, which is exactly where I want to work.

05

Knowing when not to build

I started building my own assistant for product owners. Halfway through, I realised Claude plus a few custom skills already did the job better. I stopped. Knowing when not to build is part of the job.

Let’s talk.

A role, a project or just a conversation about AI products. Send me a line; I answer every message.

matheus_castelliano@outlook.com

Open to JobsFreelanceConsulting