Indian fleet owners run their business on WhatsApp and paper. AMV had to be simpler than both.

Sankagiri moves a serious share of South India's freight on relationships, cash and WhatsApp. The businesses are established and profitable. Nobody had offered them software built for how they actually work.

• The money goes missing. Fuel, tolls and advances pass through hands before they reach any record. An approval chain, so nothing leaves without the owner seeing it.
• The people go missing. Drivers move between fleets, so an owner hiring one cannot see what he drove, what he was advanced, or whether he finished his last three trips. A per driver ledger that quietly becomes his record.
• Everything on the market was built for someone else. Enterprise logistics, with the staff and the hardware budget to match. Nothing to install, on the phone he already carries.

The category sells visibility. Operators do not lose sleep over where a truck is. They lose sleep over what a trip cost and whether the number they were given is real.

AMV is built around an approval chain rather than a map. A manager submits, an owner approves, and every rupee lands in a per driver ledger either side can open.

That ledger answers the second problem for free. Keep the money against the driver and his record builds itself, so an owner can see a man's history before he hires him instead of asking around.

In beta, with 54 vehicles onboarded and running real trips. That is a small number and it is the true one. What it buys is a working product in the hands of owners who can tell me whether it is right, before a rupee goes on growth.

The more interesting number is the one it sits next to. Sankagiri alone has 25,000 trucks and no software written for how those businesses actually run, so the category has no incumbent to displace. Get it right in one lorry town and the same product travels to the next one unchanged.

Plot the market on two axes, what they sell and who they sell it to, and one quadrant is empty. Small owners running on cash, who need control of money rather than sight of a vehicle.

Everyone else sells a box for the truck or a load to fill it. Both assume the owner's problem is the vehicle, which is the part he already understands. Subscription is the only model that works here, and it is easiest to see by elimination.

The mark is a shock absorber. Three bars at the centre for what a business is made of, its fleet, its trips, its drivers. Two covers holding them steady. Anyone who has stood in a yard knows the suspension is why the load arrives whole. Carry some of the shock, so the owner does not carry all of it.

Nothing else is allowed to add any. One accent, one job, what to do next or where you are. Warm off white, soft near black, because the light here is hard and the screens are not built for it. A number plate looks like a number plate. Made here, not translated.

A driver photographs a fuel bill and it becomes a structured expense on the right trip. The interesting part is what happens when it is not sure. It scores the match, names the fields that disagree, and hands the decision back rather than quietly deciding.

The model was never the hard part. What decides whether AI output is usable is everything around it: what good looks like written down, how the request is built, and an honest test of what comes back.

AMV was designed and built end to end with Claude. The standards were written before any of it was generated, which is the only reason it held. A model will clear whatever bar you set and no higher, so the check that sends work back is worth more when work costs nothing to make, not less.