Built with AI. Proven in production.

We have built and run software for twenty-five years. This year we rebuilt the whole process around AI — and the first product of that process runs a real business today: real cars, real customers, real money. This page is the record of it, written down while it is still hard to believe.

The old way, and its loop

Everyone who has ever bought business software knows the loop: release cycles measured in quarters, a support desk between you and a fix, and the same problems coming back with new faces. It is not because anyone is lazy. It is because the traditional process fixes symptoms and leaves the cause alive — a ticket gets patched and closed, and the class of bug that produced it survives to generate next month's tickets. Every patch adds scar tissue that makes the next fix harder, until most of a team's capacity is spent just holding still.

After twenty-five years of operating inside that loop, we knew exactly what we wanted the opposite to look like.

Then we built Ekko

In 2026 we built Ekko — a full workshop ERP: job cards, dispatch, quality control, inventory with real costing, a general ledger with VAT, HR with payroll accruals, WhatsApp built in for customers. From first build to live operation took about thirty days. It went straight into a working Dubai workshop — no pilot phase, no staging environment, one production database with real money in it from day one.

AI wrote every line of it. Our job was the part AI cannot do: knowing, from twenty-five years of operations, exactly how a business like this works, where systems break under real staff on a real floor, and what "correct" looks like on an invoice, a ledger, a job card. We test the system to failure the way real operations abuse real software; AI builds, fixes, and — this is the part that matters — proves the fix.

How it actually runs

When the workshop floor reports a problem, the fix is usually live in production within the hour. Not a quick patch — the discipline is the opposite of the old loop: the AI finds every other instance of the same class of defect across the entire system, fixes them all in one pass, and adds an automated gate that makes that class of mistake impossible to ship again. Every lesson becomes permanent, checked on every single change, forever.

The result shows in the curve. In the early weeks, half of all work was corrective — every part of the system getting its first contact with reality. But almost nothing that has been fixed has ever come back. Support load bends down while the product accelerates: in the last six weeks alone, over five hundred new features shipped alongside the fixes. In the old world those two numbers moved in opposite directions. Here they are the same motion.

The numbers

Everything below is measured from the live production system, not projected:

One more, and it is the strangest one: those 633 tables were designed, built and are maintained entirely by AI. No human has ever hand-tuned one. We watch what goes in, where it flows, and what comes out — and the system proves the rest on every change, automatically.

What this means if you're a customer

The people who sell you the software are the people who built it and the people who answer when something is wrong. There is no ticket queue between you and the fix — problems reported in the morning are usually resolved before lunch, permanently. And the product gets meaningfully better every week, because none of the capacity is being eaten by an entropy loop.

Why we're writing this down

Because in a few years, this way of building will be normal, and everyone will claim they were doing it early. This page is dated. It gets updated as the numbers move — new milestones, new products, the same discipline. That's the whole point of a record: it's written while it's still hard to believe.