Preparing the
physical world for AI.

Self-driving cars got a centimeter-accurate map of every road before they could drive. Humanoid robots are leaving the factory floor. The buildings they are headed into, where we spend ninety percent of our lives, have no map a machine can read. The most sophisticated tool running most of them is a spreadsheet.

You own or manage a property, probably the most expensive thing in your life, and when something goes wrong the experience is primitive. You call someone, explain the problem from scratch, wait for a quote that may or may not reflect reality, and hope the person who shows up knows what they are doing. Next time something breaks, the cycle starts over.

It is the same in a data hall. The drawings stop being true the day the first change is made. History lives in someone's head, or in a filing cabinet, or nowhere at all. When that person moves on, the next one reconstructs the building before the work can continue.

Every transaction around a building has two sides: a buyer and a seller, a lender and a borrower, an insured and an insurer, an owner and the crew that services it. Each side keeps its own record of the same building. None of them share one.

That is not possible while the building itself is the one thing nobody has a current model of. The record deserves the same engineering attention as the building. That is where we are putting ours.

We met in 2014 and have built together since. Before Pascal we built Ledger Investing (YC W17), an insurance technology company, and learned how much of the physical world is priced, financed and insured from documents that are already out of date.

Pascal started with the sharpest version of the problem: taking care of a home. A membership, a dedicated crew, every visit written down, live in Florida. To do that well we needed a model of each home that a technician and an agent could both read, so we built the scanner and the editor, and opened the editor to everyone.

Then the buildings AI itself lives in came to us: data centers, where a hall drifts from its drawings with every move, add and change, and where a wrong assumption costs more than any home repair.

A model of a building is rarely finished when the scan is done. It has to be checked against the building, corrected by the people who work in it, reflected in the documents the other sides rely on, and kept current when something changes.

That is why Pascal does three things and only three. Capture makes the model machines read. Edit keeps it changeable by people and agents, in the open. Service carries the plan into the building and writes reality back with every visit. A model nobody maintains is a drawing. A model that gets serviced is a record.

We are building agents that carry that context from one step to the next: from the scan to the model, from the model to the visit, from the visit back to the model, and on to the lender, the insurer and the next owner. People set the priorities, hold the licenses and make the calls.

A building affects people far beyond whoever holds the keys. The record of it should belong to the building, not to a vendor. So the editor is open source and free, every model exports to the formats the other sides already use, and every visit is on record as photos, notes and changes.

Our agents must show their work. Uncertainty must be visible. A homeowner must be able to see who came, what they did and what it changed. A data center operator must be able to trace the model back to the visit that made it so.

We will measure our progress by what our customers can rely on: homes that stay ahead of their repairs, halls whose drawings are the building, and models that are still true a decade after the first scan.

Julien & AymericCo-founders, Pascal