Salman Habib believes retailers are being asked to place too much trust in artificial intelligence, and too quickly. According to the Burq cofounder, many AI providers are seeking broad operational authority from customers from the moment their technology is integrated, without first demonstrating that it can consistently make reliable decisions. This approach, he argues, is one of the reasons retailers remain hesitant about adopting AI.

Burq is taking a different route. Its software determines how individual orders should be delivered, selecting from a retailer's own delivery fleet, gig-economy providers, parcel carriers, less-than-truckload (LTL) operators and drones. As FreightWaves explained in August, the platform acts as a decision-making layer for last-mile operations.

The stakes are particularly high because Burq's technology makes decisions that directly affect the customer's delivery experience. An incorrect recommendation can quickly translate into a failed delivery, an unhappy customer and reputational damage for the retailer.

Habib illustrated the difference between conventional generative AI applications and logistics automation in an interview with FreightWaves. While an incorrect response from Claude or ChatGPT can generally be corrected without significant consequences, assigning an order to the wrong carrier or driver can immediately create problems for both the retailer and the technology provider.

Another method of building confidence in AI

The company solidified its approach on October 13, when Salman Habib and cofounder Shaban Habib published a document titled Earned Autonomy: A Deployment Doctrine for AI in the Physical World.

The publication is the first in a series called Intelligence Notes, which Burq plans to use to detail its thinking on the use of artificial intelligence in real-world business operations.

The doctrine’s core is a simple rule: When AI is integrated into a company’s systems, it must not be given automatic decision-making power. Instead, it has to begin with no operational control and gradually gain more responsibility by proving itself on real orders.

But each step of that process still has to be approved by a human decision maker.

That idea came from a noticeable shift in the conversations we were having with potential customers, said Habib. Last year, most retailers were asking technology vendors if they had AI capabilities. More recently their questions have become much more precise: which decisions does the AI make on its own, and which decisions does it refer back to a human operator?

Burq calls the other approach granted autonomy. This model allows for an AI system to be endowed with wide powers from the moment it is hired, and its performance is measured against standards that are mostly set by its provider.

Habib is not happy about this arrangement. Vendors can do great demos and promise reliable performance, he says, but customers may have little visibility into what happened when the technology eventually makes a mistake. Without a transparent record it is difficult to identify the reason for a wrong decision and understand how it came about.

AI's four steps to taking operational authority

Burq doesn’t think of autonomy as a single capability across the company, but rather breaks operations into separate segments. Each delivery method has to build a track record before being given more decision-making power.

What’s the fuss about this difference? That’s because parcel delivery, LTL transportation, and gig-economy services are governed by different rules and limitations. “Confidence built in one category doesn’t automatically transfer to another,” says Habib.

Thus Burq uses a four-level ladder of authority on each segment:

1. Manual: The AI watches and reads the dispatchers actions but does not make operational decisions.

2. Shadow: The system records the decisions it would have taken, while dispatchers continue to manage deliveries as usual. Then its recommendations can be compared with actual human decisions.

3. Review: The AI will begin suggesting actions, but each one must be approved by a human operator before it can be executed.

4. Autonomous: Artificial intelligence is capable of making decisions on its own within the rules defined for that segment when a desired level of performance is reached.

Retailers tell us what we need to do to get the technology to the next level. For example, a company may demand that its dispatchers agree with the AI’s recommendations at least 90% of the time before the AI is allowed to make fully autonomous decisions.

The system also has automatic demotion mechanisms. If an AI-driven segment violates a mandatory operational rule, it rolls back immediately to a lower authority level with the supporting evidence attached.

Burq’s approach also differs from the widely used concept of a human in the loop. Habib says that expression requires a person to be always supervising and approving every AI decision.

But his company’s model treats human intervention as a separate step in the deployment process. The dispatchers can begin to back away from routine decision-making once they are confident enough in the system’s performance in a given segment.

Decision Ledger for AI Accountability

Transparency is another important element of Burq’s doctrine.

The Decision Ledger is a system that records every decision taken. The platform creates a record for each order, noting who or what made the decision, what information was at hand at the time, what option was chosen and what the outcome was.

Customers can also export those records to CSV format so they can review the system’s decisions outside of the platform.

“I haven’t seen any other company, whether it’s in logistics or outside of logistics, that is so open about giving customers this level of visibility into how their AI is making operational decisions,” Habib said.

Burq recommends retailers who are considering an initial deployment to begin in Shadow mode. This means companies can trial the technology without changing their current operations.

The AI quietly observes decisions and operational data across the retailer’s order management system (OMS), transportation management system (TMS) and warehouse management system (WMS). Burq is recording the choices its technology would have made, but dispatchers are still working the way they always would.

After 30, 60 or 90 days, the company can generate a report showing where its AI would have taken different decisions.

Burq has already tested its technology with one of the biggest retailers in the United States, but Habib declined to disclose its name because of confidentiality obligations.

For that first exercise, the company looked further back in the operational history, running three to six months of historical orders through its AI system. The idea was to see how the technology would choose between the retailer’s own fleet, parcel carriers, gig providers and LTL transport options.

But Burq makes an important distinction between historical testing and earned autonomy. Retailers can review past orders to get a feel for how the AI behaves, but operational authority is only achieved through performance on live orders.

Burq expands its ambitions beyond last-mile delivery

The earned autonomy doctrine is already incorporated into a product being used by customers, Habib said. Its implementation also reflects a broader evolution in Burq's positioning.

Although the company initially focused on last-mile delivery orchestration, its ambitions now cover a wider range of transportation and delivery activities.

Parcel, gig-economy services, LTL transportation and drone deliveries are all within the scope of its technology. Habib explained that Burq Intelligence considers any method of moving an order towards its destination part of the platform's operational scope.

The company is also preparing to offer customers a choice between two data-sharing arrangements.

Under the first option, a retailer's operational data will be used exclusively to support its own deliveries. Under the second, participating retailers will be able to opt in to a shared arrangement that allows them to benefit from anonymized performance information collected across Burq's wider network.

This distinction gives customers control over whether their data remains limited to their own operations or contributes to broader network-level intelligence.

For Habib, however, the immediate priority is not simply expanding the platform's technical capabilities. It is convincing retailers that they can safely begin using AI without surrendering control of their operations from the outset.

The company's strategy is therefore built around establishing trust first, with more complex technical and data-sharing questions addressed as deployments progress.

Why it matters: Burq is positioning earned autonomy as an alternative to giving artificial intelligence unrestricted operational authority from the beginning. By requiring its technology to prove its reliability separately across parcel, gig, LTL and other delivery segments, the company aims to reduce the risks associated with automated logistics decisions. Its transparent Decision Ledger, human approval stages and automatic demotion mechanisms are designed to give retailers greater control. Customers will also be able to choose whether their data supports only their own deliveries or contributes, through anonymized information, to performance improvements across the wider network.