How we work
CPQ3D is built and run by one founder with an AI agent stack. This is how that actually works, and where the line between the human and the AI sits.
Humans set the architecture. AI fills the details inside it. Everything that reaches a customer passes a human gate first.
One founder, an AI stack
Behind CPQ3D is one founder and a set of documented capabilities that handle drafting, design, analysis, and testing. Each one is a written set of rules and procedures that tells the AI how we build, not just what to make. Coding, content, configurator work, pricing, and positioning each have their own.
The instructions are the product of the work, not an afterthought. They are how a single person can run a platform that would normally need a team.
Principle before output
The model never decides the architecture. A human sets the contract: what the thing is, the rules it must obey, and the bar it has to clear. The AI fills in the details inside that frame, fast. When it gets something wrong, the rule that should have caught it gets written down, so it does not happen twice.
Disciplined AI, not vibes
Most AI-built software this year is prompt-and-ship: ask the model, post the output, let the user find the bugs. We do the opposite. Code goes through typed contracts and tests. Content goes through fixed editorial rules and a source check. Both land as reviewed changes, not direct pushes to production.
The speed comes from the structure, not from skipping it.
What AI made possible, and what it didn't
What it enabled
- +One founder maintaining an API, an admin dashboard, a marketing site, a 3D configurator, and a content library at the same time.
- +Same-week shipping. A new guide, a fix, or a feature lands in days, not quarters.
- +A full Learn library with original artwork for every guide, kept current.
- +Specialist work (pricing analysis, positioning, design) without hiring a team for each.
What it didn't replace
- −Real 3D and domain knowledge. Someone has to understand how a pergola is actually built.
- −Customer relationships. The calls, the demos, and the follow-ups are human.
- −The review gate. AI drafts; a human decides what ships.
- −Hallucination risk. The fix is not trusting the model. It is checking its output against reality.
Background
Tom Janssens has spent 20+ years in IT and has been building 3D configurators since 2015. CPQ3D is operated by Core BV, which has been running production software for over a decade.
Skeptical? Good. The real test is not this page. It is the product.
This page was drafted with AI, with a human in the loop. That is the point.