Principle
We named the studio after the part of the work nobody sees.
Usool (أصول) — Arabic: roots, foundations, first principles. The load-bearing parts. The things that decide whether what you build in year one is still standing in year five.
Seven operating principles · Published in full · Including what we refuse
- Root
- The part below ground that decides how tall the thing can grow.
- Foundation
- Poured before anything visible, judged only when it fails.
- First principle
- The thing you reason from, not the thing you arrive at.
Every firm has values. We named ourselves after ours, which means we have to live with it.
[Founder photograph]
Not a stock image. See open content items.
The founding story
In 2019 I watched a nine-month build get switched off eleven days after launch. Not because the engineering was bad — it was good — but because nobody had checked, in month one, whether the data the whole thing depended on was actually maintained. It was not. It had been hand-updated by one person who had left the year before.
The cost did not land on the vendor. It landed on the director who had championed it, who spent the next two quarters explaining a number to a board that had stopped asking politely. She had done nothing wrong except trust a plan that had never been tested against the system it described.
I decided two things that year. That I would never again work on something where the foundations had not been checked first, and that checking them would be a priced, scheduled part of the work rather than something we hoped to get to. Both of those are expensive. Both of them cost us engagements we wanted.
A rule you keep only when it is convenient is not a rule, and a rule with no name is easy to quietly stop keeping. So we named the firm after it. Usool — foundations, roots, first principles. It is a difficult thing to put on a letterhead and then not do.
[Founder name], Founder — The Usool Studio
The Seven
01 / 07
Not values. Commitments.
We refuse the engagement we cannot ship.
We have turned down work we wanted because the timeline, the data or the mandate made shipping impossible. Taking it would have been revenue; it would not have been work.
Human judgment is the last word in every consequential decision.
A model can recommend. A person decides, and that person is named in the system design.
We tell clients what they need to hear before what they want to hear.
Usually in week two. Usually it is expensive to say. We say it anyway, because the alternative is billing you for nine months of agreement.
No model in production without observability.
If we cannot see what it is doing, we do not ship it — regardless of how good the demo was.
Our work is auditable by design.
Decisions, data lineage and model versions are traceable from day one, because retrofitting an audit trail is how a good year becomes a bad quarter.
We do not build what we wouldn't deploy on ourselves.
Every pattern we recommend runs somewhere in our own stack first.
Speed without discipline is theater.
We launch loud. We can only do that because the engineering underneath is quiet.
What we won't do
The list that costs us work.
- Take an engagement where the mandate and the timeline contradict each other, and hope it resolves.
- Put a model in front of a customer without a defined path to a human.
- Staff a project with people you did not meet in the pitch.
- Use your data to train anything. Not ours, not a vendor's. It is in the contract.
- Build on a stack you cannot hire for after we leave.
- Recommend fine-tuning because it is interesting.
- Sign an NDA that prevents us from telling you a competitor has the same problem. We will never name them; we will tell you the problem is common.
- Publish a case study about you without your written approval on the final text.
If that list didn't lose you — let's talk.
Start a conversationOur stance on AI
Positions, with the reason attached.
- On autonomy
- We build systems that recommend, escalate and defer. Full autonomy is a product decision with a liability attached, and we will make you say it out loud before we build it.
- On data
- Your data trains nothing outside your own systems. We use vendor endpoints with training disabled and zero-retention terms, and we will show you the contract clause.
- On explainability
- If a decision affects someone's money, health, employment or liberty, they get a reason they can understand and contest. We will not ship a black box into those categories.
- On labour
- We will tell you honestly which roles a system changes and which it removes. Pretending otherwise makes us complicit in a rollout that fails on the floor.
- On the EU AI Act and NIST AI RMF
- We classify every system against both at design time, not at audit time. The classification goes in the architecture doc.
The Usool Pledge
- 01We will tell you what is true before we tell you what is possible.
- 02We will show you the evidence, including the evidence against us.
- 03We will name the person accountable for every decision a system makes.
- 04We will not put a model in front of your customer without a path to a human.
- 05We will build what you can maintain after we have gone.
- 06We will use your data for your systems and for nothing else.
- 07We will classify every system against the regulation before we design it.
- 08We will measure what we claimed, and publish the number we actually hit.
- 09We will recommend the smaller build when the smaller build is right.
- 10We will tell you to stop when stopping is right.
- 11We will leave the source, the reasoning and the trade-offs in your hands.
- 12We will keep this list where you can hold us to it.
[Founder 1 name], Founder
[Founder 2 name], Co-founder
Issued 01 · 2026
Ready to start with foundations?
Every engagement begins the same way: four weeks of finding out what is actually true, ending in eight artifacts you keep — whether or not we build what comes next.
The Usool Method
- Principle
- Pattern
- Prototype
- Production
- Proof