Silenzio, Bruno.
Let’s test this.

For roughly four months, Alberto, my M3 Ultra with 96GB, has been running our Autonomous Delivery House (ADH) approach. Agents work from a brief, with automated tests, independent review and decisions escalated to me when needed. The delivery work runs on Alberto, but the models we rely on are hosted in the cloud.

I tried local models on Alberto, but that setup wasn’t enough for the development work I needed. Luca and Giulia are the next attempt, with more memory and Alberto still part of the team. I borrowed the names from Luca. The veteran is already working; the newcomers have something to prove.

This series follows that next stage: local models inside an established delivery process. I’ll look at the code, the time spent fixing it, reliability and cost.

I don’t know yet whether I’ll keep both new Studios. I’ll share the failures as well as the results that look promising, and you can ask questions below each post.