It is a Tuesday afternoon in a freight brokerage outside Memphis. A rate confirmation comes in by email. It is a PDF, the way rate confirmations have arrived for decades. Twenty years ago, somebody would have printed it. Ten years ago, somebody would have typed it into a spreadsheet. Today, an AI agent built by a San Francisco company called Pallet reads the PDF, opens the customer portal, posts the update, sends the carrier a tracking link, and goes back to bed. No one printed anything. No one typed. The shipment continues.
This is the part of logistics nobody writes Medium posts about. There are no autonomous trucks here. There are no drones. There is, mostly, email. And Pallet has decided that email - plus PDFs, plus portals, plus the AS400 terminal in the corner - is the actual frontier.
The problem they saw
American freight moves roughly $1 trillion in goods a year, and most of the operational machinery behind it still runs on documents written for a fax machine. Bills of lading. Rate confirmations. Customs paperwork. Carrier portals, each one slightly different, none of them with a working API. The industry's response, for decades, has been to hire people - thousands of them - to retype things from one system into another.
Co-founders Sushanth Raman and Andrew Spencer noticed this from an unusual vantage point. They were not freight people, exactly. They were engineers at Retool, the low-code platform that companies use to build internal dashboards. They watched logistics teams come in looking for help with the same workflows over and over: extract this field, route it to that screen, update the customer.
They also had family in the business. Raman's grandfather was in shipping. Spencer's father runs engineering at MercuryGate, one of the larger transportation management vendors. The pair already knew the part most outsiders learn the hard way: logistics is not a software problem with a thin paper layer on top. It is a paper problem with a thin software layer on top.
The founders' bet
The polite version of automating freight back office is RPA - the click-recording bots that have been sold to enterprises since the early 2010s. They work, in the sense that a typewriter works. They break the second a portal updates a button or a PDF arrives in a slightly different layout, which in this industry is roughly every Tuesday.
Raman and Spencer made a different bet, one which now looks obvious and in 2020 did not. They bet that large language models would soon be good enough to do this work the way a human operator does it - by reading, judging, and clicking - rather than the way a script does it. The product they built, CoPallet, leans into that. It is less a workflow engine than what the company likes to call an "AI workforce," which is marketing language that becomes more accurate the longer you watch the thing run.
Pallet's early investors took the bet seriously. Bain Capital Ventures led the seed. Activant Capital and Bessemer joined for the Series A in October 2024, an $18M round. Last May, General Catalyst led a $27M Series B, bringing the total to roughly $48 million. None of this is generationally large by AI-funding standards. It is, however, an unusual amount of capital to point at order entry.
The file on Pallet
The product
CoPallet is best understood as a collection of digital co-workers, each of which has been given access to the same systems a freight operator has access to - email inbox, customer portal, TMS, WMS, sometimes a 1980s-vintage AS400 mainframe - and instructed to perform a specific job. Quote requests, order entry, status updates, document parsing, proof-of-delivery retrieval. The system claims 97% accuracy with built-in guardrails, which in practice means a human gets pinged on the edge cases and only the edge cases.
Two newer pieces deserve mention. Forge lets logistics teams build their own agents, an admission that no central R&D team can keep up with every flavor of warehouse workflow in North America. Parallel Agents runs many agents at once against a single shipment, which is the kind of feature that sounds incremental until you have watched a single multi-leg international load and counted how many separate things have to happen to it.
// Workflow reductions reported by customers
Pallet, by the numbers customers will actually quote
Source: customer statements published by Pallet and trade press. Your mileage will vary - some of these warehouses still print things.
Five years, in five dots
The proof
Press releases are easy. Customer logos are harder. Pallet has both, and the second list is the more interesting one. STG Logistics, a large North American intermodal player, reports 95% of inbound orders fully automated. Mallory Alexander, a logistics company that has existed since 1925, claims a 3x throughput jump per operator and 100% accuracy on import filings. Prism Logistix attributes a 10% margin improvement to the platform. Eassons Transport Group went live in 40 days, which anyone who has implemented enterprise logistics software will recognize as suspiciously fast.
There are larger names quietly using the product as well - Knight, Swift, Lineage, Rinchem, USPACK, Estée Lauder. Not all of them want to be the public face of "we replaced a back office team with AI," for reasons that are partly cultural and partly union-shaped. Pallet does not seem to mind. The company has been studiously quiet about itself for most of its life, an unusual strategy in San Francisco and, in this category, probably the correct one.
Seed, Series A, Series B. Investors include General Catalyst, Bain, Activant, Bessemer.
Engineering-heavy, customer-embedded. Engineers visit warehouses in person, which is rarer than it sounds.
Yes, that AS400. Pallet's agents will happily talk to a mainframe older than most of the team.
The mission
Pallet's stated mission is to automate the costly, repetitive work that runs modern logistics so that freight teams can move more goods, with fewer errors, at lower cost. The unstated mission - the one you have to squint at to see - is harder. The company is trying to make AI feel boring to an industry that is allergic to hype, by shipping software that solves a Tuesday-afternoon problem reliably enough that the warehouse manager forgets it was ever AI in the first place.
This is not a small bet. Freight has eaten several decades' worth of digital transformation projects and asked for seconds. The graveyard of TMS rip-and-replace efforts is large and well-tended. Pallet's wager is that AI agents do not require the rip-and-replace at all - that they can sit on top of the broken systems, work the way humans worked, and quietly turn the lights off in the back office.
Why it matters tomorrow
The freight industry employs millions of people. A large share of that headcount sits behind a screen typing things from one window into another. Pallet is not the only company building agents for this work, but it is one of the few that has gone in deep enough to handle the parts that do not fit on a slide - the bad PDFs, the carrier portals that change weekly, the mainframes, the customer who insists on email. Whoever wins this category will not win it with the best demo. They will win it with the most patience.
It is also worth noting what this does and does not displace. Pallet's customers tend to report that they keep their teams and grow into the available capacity rather than firing people. That is a quieter, more interesting story than the one usually told about AI and work, and it is happening - of all places - in trucking.
Back to Memphis
It is still Tuesday afternoon. The rate confirmation has been processed. The portal has been updated. The driver knows where to be. The operator who five years ago would have handled this load is on the phone instead, working an exception that actually needs a human - a customs delay, a damaged pallet, the kind of thing that requires judgment and a small amount of swearing. The brokerage is moving more freight than it did last year, with about the same number of people. Nobody printed anything. Nobody typed.
This is what Pallet has done, and it is what makes the company worth watching. The dramatic version of AI replacing work has been on a slide deck for a decade. The undramatic version is happening in a freight office outside Memphis, and the company behind it lives quietly in San Francisco, sends engineers to warehouses, and refuses to oversell.