AI + People, Part 2: The Partnership

In our last AI+people article, we walked through a brief history of human-computer interaction (HCI). The timeline took us from the introduction of paper punch cards through the creation of the intuitive graphical user interface (GUI). That history was one defined by a clear, transactional dynamic—the human was the operator and the computer was the tool.

You input BOL details, it saved them.

You clicked submit, it transmitted the forms to CBP.

The truth of the current moment is a bit more complex than it looks at first glance. In reality, we’ve crossed a key threshold and are no longer simply building better tools—we’re building partner agents. This shift has triggered a wave of anxiety across the logistics world, with the prevailing narrative being one where the artificial intelligence (AI) agent is capable of planning and executing on its own, reading invoices, classifying goods, and drafting entries. This narrative ends with the role of the broker being diminished past the point of no return.

This view fundamentally misunderstands the true trajectory of both our industry and this technology. “Automation” is not synonymous with “replacement.” In other words, we’re not aiming for a workerless state of business; we’re aiming for a “Great Partnership,” where “human-in-the-loop” (HITL) architecture allows the power of computers to ingest documents at scale and recognize patterns at speed to combine with the human capacity for regulatory judgment and client service. We aren’t removing experts from the equation—we’re liberating them to do the high-value work that only experts can do.

From “operator” to “partner”: the evolution of the loop

To understand where we’re going, it’s important to first look at how “the loop” has changed. In traditional freight forwarding, the loop was open-ended and fully dependent on human initiation. A customs entry writer received an email with a commercial invoice PDF, opened their TMS, and manually keyed in the shipper, consignee, and line items one by one. The software was passive, waiting for the keystrokes, processing the data, and returning a result.

Today, AI agents are closing that loop. Unlike a passive optical character recognition (OCR) tool, an AI agent doesn’t just scan text and convert it to digital data. Rather, it extracts data from the tangle of unstructured emails and invoices in different formats that it ingests, maps it to the right fields in your existing systems, cross-references it against HS code databases, and drafts the entry for you.

It’s this seemingly newfound autonomy that unsettles some observers. If the agent software can “read” the invoices and “write” the entry, does it still need people?

The short answer is, yes, with the caveat that the relationship has shifted. In a properly designed HITL system, the broker hasn’t left the loop—they’ve moved into the center of it. Brokers are transitioning from being data entry clerks to being architects of the shipment lifecycle. The goal of integrating AI into forwarding operations is not to remove the entry writer. The goal is to elevate them from processing paperwork to managing exceptions and developing strategy.

The division of labor: speed vs. significance

To build a successful partnership as described above, we need to be clear about who does what. Collaboration like this partnership works best when you can play to the respective strengths of both parties. 

The agent’s domain: speed and automation

AI agents excel at tasks that are computationally expensive but cognitively repetitive. They thrive on speed, scale, and pattern recognition. Consider the bane of many brokers’ days: the drudgery of re-keying data from PDFs.

Agents can be deployed to scrape incoming emails for quote requests, extract line items from 50-page packing lists, and reconcile accounts payable invoices against accruals. They can process thousands of lines of data instantly, flagging missing weights or inconsistent values that would take a human hours to verify manually. By handing this job to an AI agent, we are not “automating jobs away,” we’re automating the parts of the job that burn out your best employees and lead to human error. 

The human’s domain: criticality and creativity

This automation creates gaps that are best filled by human abilities. The human contribution is defined by contextual understanding, regulatory nuance, and empathy for clients.

First comes critical thinking. An agent can tell you what the commercial invoice says, but it struggles with why. An agent may classify a new product based on a keyword match, but a seasoned expert will know that “safety boot” has a different duty rate depending on the exact material composition and intended end-use. The data point is the same, but the regulatory context completely changes the decision.

Second is exception management. AI models are trained on historical datasets, in other words, they’re mirrors of the past. When a unique situation emerges, a port strike, a sudden sanction change, or a frantic client needing to reroute a container mid-ocean, for example, the agent will likely fail. Humans are necessary to bridge these gaps, finding creative solutions for edge cases that historical data can’t predict.

In this partnership, the agent provides the velocity by churning through the documents, while the human brings the vector by ensuring this speed is directed toward a compliant and profitable outcome for all parties.

Designing human-in-the-loop architecture

Great partnerships don’t happen by accident. They require intentional system design that prioritizes our values of transparency and control here at Reform. We can’t simply plug these models into our operations and hope for compliance. We must design HITL architectures.

Transparency vs. the “black box”

A major risk with advanced AI is the “black box” problem, where the system produces an output without revealing its reasoning. In customs and forwarding, this is dangerous. You can’t explain a misclassification to CBP or HMRC by saying, “the AI did it.”

HITL systems are designed for explainability. The agent shouldn’t just pick an HS code, it should present a case for why it picked that code. It should be able to highlight the specific description on the invoice that led to its suggestion and show a confidence score. This allows the human partner to validate the logic, not just rubber-stamp the result.

The need for “good friction”

Efficiency is usually about removing friction, but in the age of AI, we need to reintroduce the idea of “good friction.” We must incorporate pauses in the workflow where the agent cannot proceed without human sign-off. This is critical for high-stakes decisions. An agent can be allowed to draft a quote or queue an ISF filing, however, before transmitting a final summary entry to CBP or releasing cargo, the system must force a pause. It must require a licensed customs broker to review the work and take ownership of the final decision.

The oversight mindset

This shift requires a change in how teams are trained. Operations staff must be upskilled to move from “doing the work” to “auditing the work.” They must become critical reviewers of AI outputs. This model also reinforces a core value: accountability belongs to people, not code. The broker’s license is on the line.

When a penalty is issued, names are named, and a human must be held responsible. The HITL architecture ensures that there’s always a human face behind compliance, which is essential for maintaining trust with both authorities and customers alike.

Conclusion: the future is more human, not less

If we succumb to the fear of replacement, we risk stalling the very progress that could save our industry from margin compression and burnout. But if we embrace the Great Partnership, the prize is significant. We don’t just get cheaper operations, we get better service. We get a forwarding team that spends more time advising clients on supply chain strategy and less time copying and pasting from PDFs to systems of record. We get a compliance process that is proactive rather than reactive.

The promise of AI for Reform customers is not about creating a robotic workforce. It’s about using technology to strip away the robotic parts of human jobs—the typing, filing, checking, and rechecking—so we can return to the core of logistics: solving complex problems for people moving goods around the world.

‍