01 · The same job, six times.
Take a single electrical circuit from a new-build flat. A 32A radial feeding the kitchen sockets. That one decision — one circuit — will be described, typed, and re-typed across at least six systems before the job is done.
It appears on the design drawing, so a designer types it. It appears in the estimate, so an estimator types it. It appears on a purchase order, so someone keys it into a supplier portal. It appears on a test certificate, so the electrician writes it again on paper, then someone transcribes it digitally. It appears in the O&M manual and the handover pack. And when the final payment application is built, the same circuit is described a sixth time.
Every step is a person. Every step is a chance for error. Every step is typing that doesn't add value.
02 · Digitised is not connected.
The last twenty years of construction software answered the first question — can we make this digital? — without ever answering the second: does one stage talk to the next?
You can now produce an EIC on a tablet. You can draw on a tablet. You can estimate on a tablet. The work used to be on paper. Now it's on a tablet. And the same circuit is still typed six times, because none of those tablets know the others exist.
03 · One live substrate.
The model Prekon is built on is narrow, and we've not changed it since we started: one live data object per job, with every output — drawing, estimate, cert, O&M, payment application — rendered from the same source.
You don't produce a cert. You don't produce an estimate. You describe a job once — sockets placed on a drawing, circuits inferred, cables costed, labour allocated — and the outputs are views over the same object. The cert is the drawing, re-rendered as BS 7671. The estimate is the drawing, re-rendered as money and hours. The O&M is the drawing, re-rendered for a facilities manager.
This isn't a new insight — architects have thought in these terms since BIM arrived in the 2000s. The news is that the cost of actually building a unified substrate has just collapsed. LLMs do the translation work that used to require a full integration team. Vector PDF tooling has caught up. Firestore and agents and cheap compute did the rest.
04 · Agents as the interface.
The second half of Prekon's bet is that agents are the right interface for this substrate — not forms, not dashboards.
A contractor doesn't want to learn a new UI. They want to say: "Add a shower circuit to the master bathroom, 10mm, on its own RCBO, and update the cert." That's six database writes, three calculations, one regulation check, and a PDF regeneration. An agent does it. A form takes ten minutes and still misses one field.
Agent-first means designing the substrate for agents to drive, not designing an app that has an agent bolted to the side. That's the difference between Prekon and the next wave of "AI-assisted" construction tools.
05 · What we're building, honestly.
The full single-substrate vision — design through to digital twin — is a decade of work. We're not claiming it's built. We're claiming it's the north star, and we're working from the cert side inwards.
Certificates work today. Remedial estimating is in active build. Drawing and spatial extraction are live R&D with working vector extraction but not yet ready for contractors. The digital twin is a thesis piece — directional, not promised.
If this is the bet you want to be close to, register interest. If you want to watch from a distance, follow the build publicly. Either way — the design, the estimate, the cert, and everything in between, are becoming one thing.