Speed
What actually got faster
Not generated lines. Working capabilities, repeatable artifacts, and elapsed time to something I could deploy.
Après Studio
I built something.
Then I tried
to break it.
Après Studio started with a simple question: what can AI actually build on a real enterprise commerce stack?
I wanted to find the limits on about every damn thing.
The rule for the first build was simple: AI writes the storefront. I do not touch the code.
I gave it OCAPI, SCAPI, WebDAV, MRT, and an empty starting point. The work also includes catalog tooling, content, integrations, deployment, testing, change, and recovery.
I keep the interventions and failures visible. A finished-looking storefront does not tell you how much work remains.
01 /Evaluation
I’m keeping score.
Speed matters. So do the mistakes, token burn, rework, and hidden human context.
Intervention
Where I took the keyboard back
Missing context, bad assumptions, platform constraints, defects, rewrites, and decisions that still needed experience.
Ownership
Whether another team could live with it
Testing, observability, safe change, failure handling, and enough context that the whole thing is not trapped in my head.
02 /Boundary
What counts.
- 01
Working
The system does the intended job against the platform, data, and integration constraints.
- 02
Repeatable
The process can be run again without relying on hidden context or a lucky prompt.
- 03
Ownable
The result can be understood, changed, tested, deployed, and recovered by a real team.