From the web to governed agentic AI
My path runs from the web and campaigns to governed agentic AI, and my method is simple to state: the agents do the work and I make the decisions.
I'm Francisco Ortiz Aguilera, a senior software engineer and Creative Technologist, and I sign my work as Baronscript. My path runs from the web and campaigns to governed agentic AI, and my method is simple to state: the agents work and I decide.
A craft that has kept changing ground
I have worked in a creative agency, on websites and CMSs, on campaigns, on bespoke platforms, on data and, most recently, on agentic AI.
I started out as a freelancer in 2006, with projects for town councils, the University of Granada, the CSIC and businesses. From August 2016 I worked as a developer at Shackleton Group, an independent creative agency that Accenture acquired in 2019. Since then I have been a software engineer at Accenture, now Accenture Song, working remotely on full-stack development for clients.
I'm a senior full-stack engineer: I work in PHP, JavaScript/TypeScript and Python, with Laravel, Angular and Docker. Design is part of the craft too: I have advanced training in UX/UI through the Front-End track of the Interaction Design Foundation (IxDF).
An agentic AI laboratory
Since September 2026 I have run my own agentic AI laboratory on my homelab, alongside my work at Accenture Song, using Python and MCP. There, the method becomes concrete rules.
Separation of duties. Whoever creates a change never reviews or approves it, and a person approves the product, architecture, implementation and release milestones.
Protocol and permissions. I have designed my own communication protocol between agents, with a common message envelope, a trace model and an event catalogue. Each agent works with scoped permissions and under a policy for models and data egress.
Architecture. More than a dozen subagents share a common template. The pipelines build software as graphs of parallel tasks, with deterministic gates and four independent reviewers per task.
Defensive security. I have designed my own defensive security specialist, ASE, which audits only authorised assets, with reproducible evidence, an independent reviewer and human approval.
Local AI. I also work with local AI: open models served with Ollama on two instances behind a load balancer, which I use for prompting, embeddings, semantic search and reranking.
Applied to my own website
This website and its copy were produced in that way: with agents and with my approval.
What you should know
Everything I say here is published by me on baronscript.com and has been reviewed against evidence by my research system; there is no third-party verification. On the projects and experience pages I name the brands I have worked with, but I don't share internal details of their projects.
Let's talk about your project: baronscript.com/contacto