AI agents, built around the constraint

Every industry has one rule that decides what software is allowed to do on its own. In Dubai real estate it is a permit. In UAE ecommerce it is cash on delivery. In logistics it is that the worst failure never announces itself. These pages start from that rule rather than from a feature list, because it is the rule that determines the design.

The method, whatever the sector
1Find the step that is irreversible or regulated, and make that the one the agent hands to a person.
2Find the failure that produces no signal, and give it a clock.
3Find the two systems that disagree, and decide which one is the record of truth before writing a line of code.

Which industry are you in?

Four industry pages so far. Each names the constraint it is built on, the systems the agent touches, and where it has to stop. Open one.

Trakheesi sits in front of everything else. A listing without a valid advertisement permit number has no lawful route to Bayut or Property Finder, so the whole design turns on a refusal rather than a capability: the agent has to be unable to publish, not merely reluctant to. Everything after that, chasing enquiries and keeping the same unit consistent across two portals, is ordinary work.

BayutProperty FinderEjari datesCRM

Read the real estate page →

The agent does
Portal enquiries answered, listings kept consistent across portals, Ejari renewal dates tracked.
Where it stops
Nothing is published or edited without a valid permit number on the record.

Cash on delivery puts days between the handover and the money. Until a courier remittance has been matched to the order it settles, the revenue figure, the refund and the stock count are all provisional, and an agent that treats any of them as final is acting on something that has not happened yet. The matching has to come first.

StoreCourier remittancesStripeStock

Read the ecommerce page →

The agent does
Orders confirmed before dispatch, remittances matched, stock kept honest across channels.
Where it stops
A refund is held until the courier has actually remitted.

Every other automation on this page waits for something to happen. Here the expensive case is a shipment that does nothing: no webhook, no status change, no error, just silence where an update should have been. You cannot trigger on that. It takes a clock, and a threshold set per lane, because a normal gap on one route is a stuck container on another.

TMSCarrier APIsCustoms docsSlack

Read the logistics page →

The agent does
Shipment status, proof of delivery and customs paperwork chased on a schedule.
Where it stops
A lane past its threshold goes to a named person with the last known event attached.

The line here is drawn in law, not in preference: the agent stays on the administrative side of it and never touches a clinical judgement. The harder constraint is architectural. Federal Law No. 2 of 2019 keeps health data generated in the UAE inside the country, so residency decides where the build can run before a single workflow is designed.

Clinic systemInsurer eligibilityIntake formsWhatsApp

Read the healthcare page →

The agent does
Scheduling, insurance eligibility, intake and follow-up.
Where it stops
Anything clinical, and anything that would move health data out of permitted storage.

What if your industry is not here?

These four are the ones written up so far, and there is no longer list waiting behind them. These four came first because each has a constraint specific enough to explain in public without inventing anything.

The method does not change with the industry, and that work is the same whether the systems are property portals, a store or a transport management system.

If you want the service view instead, all services lists what we build, the use cases page starts from the problem, and how we work covers scope, fixed price and what happens after launch.

What rule does your business run on?

Name the rule your operation runs on, the one a person currently enforces by remembering it. We will tell you whether an agent can be trusted with it, where it would have to stop and ask, and what the build would cost.