Highly specialised agentic workflows, in action.

Everything on this site — the production studio, the pipeline behind it, the guide video explaining it, even this page — was built in a running conversation between a human architect and an AI system. The human typed no code on this project. That is not the achievement it sounds like, and it was never "AI did it all": it worked because every output was reviewed, questioned and quality-assured by someone who has spent more than twenty-five years building software by hand.

And the workbench itself is the deeper half of the experiment. This is not another app with AI features bolted on. The studio is an agentic workflow — a scriptwriter, a slide designer, a voice director, a publisher, each an agent with a specialised job and a defined hand-off — dressed up for human interaction. The pipeline lights, the revision boxes, the play buttons: that interface exists so that directing agents stops hurting. No prompt engineering, no chat window, no guessing what the AI did or what it cost — you see stages, you say in plain words what should change, and the spend sits on the project in plain sight. Call it agentic pain relief.

Experience is the invisible ingredient. A career from senior backend engineer on Azure and Kubernetes, through technical project lead and architect, to engineering manager and senior software architect shaping agentic development workflows — that is what compresses into one-sentence direction. Corrections like "that chart is hard-coded in the renderer — that's bad, the model should generate it" steered this project at decisive moments, and only an experienced eye makes them at a glance.

The domain is almost beside the point. Voice-overs and slide decks made a vivid showcase, but the method transfers to any domain: break the vision into stages an agent can own, keep a skilled human in the loop at every one of them, make every result verifiable and every cost visible — and push the AI to the edge of what it can do, in a human-first, friendly way. Every feature here began as a sentence like "we should be able to adjust timing in the slide show" and became working, tested, deployed software — usually within the hour, with the spend for every step accounted for on the project itself.

If this way of working speaks to you — for your domain, your team, your problem — he'd like to hear from you.