Salman Habib thinks AI vendors are asking retailers for too much, too soon. The vendors pitching them so far want full trust from day one, the Burq cofounder said, and that is why many retailers are still stuck deciding whether to use AI at all.
Burq’s software decides how each order gets delivered, choosing among the retailer’s own fleet, gig providers, parcel and LTL carriers, and drones, the role FreightWaves described in August as the last-mile decision layer. That puts its AI where a mistake reaches a customer’s door.
“If Claude or ChatGPT gets a sentence wrong, there’s no hard feeling. You tell it to fix it and then that’s pretty much it,” Habib told FreightWaves. “Us getting an order wrong, us dispatching something to the wrong carrier or to the wrong driver, is an angry customer, and that looks bad on the retailer, that looks bad on us as a company.”
Habib and cofounder Shaban Habib laid out their answer Oct. 13 in “Earned Autonomy: A Deployment Doctrine for AI in the Physical World,” the first entry in a series Burq calls Intelligence Notes. The doctrine holds that AI should start each part of a business with no authority and earn more only on evidence from live orders, with a person signing off at each step.
What buyers ask AI vendors now
Habib said the push for the doctrine came from his own sales meetings.
“Until last year it was ‘Do you have AI? Do you have AI?’ and everyone was asking that question, right?” he said. “And suddenly it changed to ‘What does your AI decide today, and what does it hand back to a person …?'”
The doctrine calls the pitch it argues against “granted autonomy”: full authority on integration day, with evaluation left to the vendor’s own benchmark. In Habib’s telling, the vendor gives an “amazing demo,” asks for the whole operation, and promises the AI won’t make a mistake.
“Until it makes a mistake, and then there’s no way to look at what happened, what went wrong, and why did it go wrong,” he said.
Authority earned by segment
Burq’s version splits the operation into segments, such as delivery mode, so trust earned on one doesn’t carry over to the next
“Burq AI will earn its trust for parcel separately than it does for LTL, than it does for gig, because the operations, the mechanics of these could be very different,” Habib said.
Each segment climbs four rungs on what Burq calls an authority ladder: Manual, where the AI only reads dispatchers’ work; Shadow, where it logs the call it would have made beside theirs; Review, where it proposes and a person accepts; and Autonomous, where it acts on its own within the segment’s rules.
The retailer sets the bar for each promotion. In Habib’s example, a retailer could require its dispatchers to agree with the AI’s call 90% of the time before granting full autonomy. Demotion is automatic: breach a hard rule, Burq’s site says, and the segment “demotes itself immediately, evidence attached.”
Habib said Burq avoids the industry’s usual term for this arrangement, human in the loop.
“Human in the loop suggests that there’s always a human being sitting there rubber-stamping all the decisions,” he said. “We don’t need that. We just need human to be a stage in the process.”
Once dispatchers are confident in the AI’s calls on a segment, he said, they can step back and let it run.
A ledger of every call
Every decision lands in what Burq calls the Decision Ledger, one row per order showing who decided, what was known, what was chosen, and how it turned out. Customers can export it as a CSV file.
No other company he has seen, in logistics or anywhere else, openly shows customers how its AI made its decisions, Habib said.
For a shipper weighing a trial, Habib pointed to Shadow mode as the starting point.
“Your operations don’t change. Everything stays as it is,” he said. “Our AI silently just reads your decisions, your operations, all of it. Your OMS, your TMS, your WMS, all of it.”
The result is a report on what Burq would have done differently over the first 30, 60, or 90 days.
Burq started with one of the largest US retailers a step earlier, running three to six months of its past orders through the AI and reporting how it would have chosen among the retailer’s in-house fleet, parcel, gig, and LTL options.
Under the doctrine, looking at past orders can show a retailer what the AI would do, but the AI only starts earning authority once it runs on live orders. Habib did not name the retailer, citing confidentiality.
Past the last mile
The doctrine is already built into a product customers are using, Habib said, and it has widened Burq’s scope beyond last-mile delivery.
“It could be gig, it could be parcel, it could be LTL, it could be drone,” he said. “… To us, any form of movement, a way for an order to get delivered, is part of Burq Intelligence.”
Customers will be able to choose between two data arrangements, Habib said: one where a retailer’s data serves only its own deliveries, and one where retailers can opt in to benefit from anonymized performance data across the whole network.
The first thing Burq has to solve for, he said, is to “give them that trust to get started with. And then these technical questions of course will follow along.”
Why it matters: Burq is pitching its customers will be able to choose between two data arrangements. One where a retailer’s data serves only its own deliveries, and one where retailers can opt in to benefit from anonymized performance data across the whole network.
