How I built a multi-agent studio system to ship my own products
Before I ship my own products, I built the system that runs the studio. Hyde Studio is a multi-agent setup that holds my place, tracks experiments, and makes sure things actually get finished, not just started.
I build websites and AI products, I do the marketing to get them found, and I have decades in specialist retail and ecommerce. Many people who build cannot market, many who market cannot build, and very few have lived in the retail and ecommerce world they are selling to. That combination is what Hyde Media is, and I have just built a system to point it at shipping my own products, not only client work.
This is the first entry in a build-in-public log. The plan is simple: show the work as I go. What I build, how it is put together, what makes money and what does not. This first post is about the system itself, because before I ship the products, I built the thing that runs the studio.
What it is
Hyde Studio is a multi-agent system that runs my product studio. It is a small team of specialist AI agents, each with one job, orchestrated from a single command line. It plans experiments, holds me to actually finish them, writes them up, watches the market and the AI frontier, keeps the books, and keeps me honest. It is a sibling to a system I already run day to day for ebikeist, a premium e-bike retailer I do the content and digital work for, and it forks the patterns that already earned their place there.
Why build a system instead of just building the products?
Two reasons. The first is that I want to ship my own things, not only service work. Service work pays, but it does not compound. A product you build once and sell many times does.
The second is harder to admit. For me, the difficult part of building is rarely ideas. It is finishing, and not losing the thread. Any solo builder knows the pattern: you go quiet for a fortnight, life gets busy, client work picks up, and when you come back you cannot even remember where you stopped. So the core of this system is a spine that holds the place. I can pause for two weeks, open it on a Monday, and it tells me exactly where everything stood and what is next. The thing that used to kill projects, falling off and not knowing where I was, becomes very hard to do by accident.
How is it put together?
The design rule is strict: every agent either scaffolds a weakness of mine or multiplies a strength. If it does neither, it does not get built. Small teams win, because every agent is one more thing to maintain and trust.
Underneath sits the spine:
- A status command that outranks my memory. I do not get to claim what day it is or what state things are in. A script walks the project, checks the clock, and tells me. If memory and the script disagree, the script wins.
- A findings tracker. Every piece of work with an owner and a verb becomes a numbered entry. Nothing important is allowed to live only in my head, or buried in a document nobody reopens.
- File-based memory. When something goes wrong, or I confirm a way of working, it gets written down as a durable rule so it is not relearned the hard way.
On top of that spine runs the team. A Pipeline holds the queue of experiments, each with a quiet question riding alongside it: who would pay for this, and how soon. A Shipper owns the last ten per cent, and refuses to call something shipped until it is deployed and reachable, written up, and one real person can actually try it. Without that, "ship it" is just a vibe. Around them sit the rest: a Publicist that helps me write the work up, a Scout and a Frontier-watch that keep my market and my edge current without the hype, an Accountant that keeps the money straight, and a Coach that asks me how the day actually went.
What is next
The first build is a client app for a personal trainer, made as a design partner rather than a paying client, then generalised into something the trainers who will pay can use. Alongside it, a piece of work I have already proved once: helping specialist retailers get found by AI search, the way I did for ebikeist's buyer's guide.
I will write each of those up as they happen, the wins and the misfires both. If any of it is useful to you, or you are building something where building, marketing and a real understanding of a sector all have to meet, I would love to hear from you.