Antonio Ramos Torres AI compounds architecture.

Software Architect · Lead Software Engineer at Vibia · Barcelona

For over 10 years I've designed maintainable systems with DDD and hexagonal architecture. Today I build end-to-end agent factories that let my team ship an order of magnitude faster, with better quality and better performance.

Antonio Ramos Torres, software architect in Barcelona
designing systems
10+ years
engineers on the team I built from scratch
~10
of magnitude faster delivery with AI
An order
games published, 17 of them in 2026
18

Agent factories, end to end

I don't use AI as autocomplete. I design the whole system: which agents are needed, what context they get, how they coordinate, how their output is measured and how it reaches production. It works because it sits on an architecture with clear boundaries and tests that don't lie.

  1. Step 1 Context What the agents know about the system: boundaries, conventions and tests.
  2. Step 2 Agents and skills Specialised, reusable pieces built on Claude Code.
  3. Step 3 Orchestration How they split and chain the work between them.
  4. Step 4 Evaluation How their output is measured before it is accepted.
  5. Step 5 Production How it reaches production, within the boundaries of the architecture.

Agents and skills for the team

I started the adoption myself and extended it to the whole team with agents and skills on Claude Code: the factory does the repetitive work, people make the decisions.

Proven outside work too

I use the same method at DevApplications, my studio: 17 games shipped in 2026, built with AI from start to finish.

From prototype to production

Prototypes that don't stay demos: they're born inside the architecture, so the jump to production is short.

Quality and performance, not just speed

Faster, yes, but with more tests and better performance: the factory works inside the boundaries the architecture sets.

Case studies

From coupled microservices to a hexagonal monorepo

Problem

A legacy PHP/Symfony stack with coupled microservices and business logic tangled with infrastructure.

Decision

DDD and hexagonal architecture in a monorepo; modular React + Next.js frontend and Node.js backend; a standalone API built with TDD.

Result

A dockerized, testable, maintainable codebase that both people and agents now work on.

A team from zero to ten

Problem

The team had to be built alongside the platform.

Decision

Hire for judgement, not stack, and mentor continuously.

Result

~10 engineers, several of whom grew from junior to senior contributors.

AI, from one person to the whole team

Problem

AI used now and then gives results now and then.

Decision

Start with myself, turn what worked into shared agents and skills, and roll it out to the team.

Result

We ship an order of magnitude faster, with better quality and better performance.

Experience

  1. May 2022 - Present · 4 years 6 months

    Lead Software Engineer / Software Architect · Vibia

    Barcelona
    • I design and build end-to-end AI agent factories, and took their adoption from myself to the whole team with agents and skills on Claude Code; we ship an order of magnitude faster, with better quality and performance.
    • I take prototypes to production quickly by building them inside the architecture from day one, not alongside it.
    • Defined and implemented DDD and Hexagonal Architecture, replacing coupled microservices with a clean, testable monorepo.
    • Replaced the legacy PHP/Symfony stack with a modular React + Next.js frontend and a Node.js/Express backend, with a standalone API built with TDD.
    • Built the team from scratch (~10 engineers): hiring, mentoring and growing juniors into senior contributors.
  2. Feb 2021 - May 2022 · 1 year 4 months

    Senior Software Developer · Signa Sports United

    Remote (Europe)

    Modernised a legacy Symfony app: DDD to decouple business rules from the framework, Docker, CI/CD, TDD and Cypress.

  3. Oct 2020 - Feb 2021 · 5 months

    Software Developer · Mayordomo Smart Points

    Barcelona

    Symfony platform: backend and API tests, working side by side with the CTO.

  4. May 2019 - Oct 2020 · 1 year 6 months

    Software Developer · PromoFarma.com

    Barcelona

    Marketing microservice in PHP/Symfony, MongoDB and Docker with CQRS and high test coverage.

  5. Jul 2017 - Apr 2019 · 1 year 10 months

    Developer · LaComunity

    Barcelona

    Laravel/MySQL monolith, fast business-driven iteration.

The lab: 18 published games

DevApplications is my indie studio. It started in 2014 with an Arkanoid that has now passed half a million downloads; in 2026 I've shipped 17 more games, built with AI from start to finish. There's craft behind each one: solvers that verify every puzzle has exactly one solution, AI opponents whose levels are measured against optimal play, localization and a website with a page for every game.

games
18
in 2026
17
downloads (Arkanoid)
500K+
on Google Play
4.4★
See the games at devapplications.xyz
Arkanoid Classic screenshot

Open source

Let's talk

I'm not looking to move, but I'm always up for a conversation about architecture, teams and applied AI.