hmdeveloper.com

Consulting

One-on-one sessions to move past improvised AI-generated code and build an agent workflow that holds up on a real project.

Who it is for

  • Developers new to AI who already use Claude Code, Cursor or Copilot and want to understand what happens underneath
  • People building by vibecoding who get stuck as the project grows: context blowing up, the agent rewriting what already worked
  • Teams that want to standardize how they use agents before everyone invents their own process

What we cover

  • Skills: when to create one, how to structure it, and why most of them fire at the wrong moment
  • Subagents: splitting responsibility, routing each task to the right model, and deciding where review belongs
  • MCP (Model Context Protocol): connecting your own tools to the agent, with writes behind a confirmation step
  • Context: organizing it in layers so session cost does not climb with the size of the codebase
  • Workflow: worktrees, pull requests, automated review, and what is better left manual

How it works

  1. 1You describe your context and what is blocking you in the form below
  2. 2I reply saying whether I can help. If it is not my area, I say so right away
  3. 3One hour over a call, recorded, with a written summary of what was decided

I started offering this publicly not long ago, so early sessions are priced below what I intend to charge later. If your problem is not something I solve well, I will say so before we book.

Tell me about your case

Claude Code and AI agents consulting | Hugo Minari Diniz