The Man Who Would Automate the Ocean

Ryan Petersen does not talk like a man running a $2 billion logistics company. He talks like someone who has spent a great deal of time thinking about coins, prophets, and the infinite hunger of the human soul — and who happens, almost incidentally, to move cargo across the planet for a living. Seated across from the hosts of The Light Cone, Y Combinator's podcast about the frontier of startups, the Flexport founder opens with a claim that would sound like bravado from anyone else: that artificial intelligence can make the price of shipping anything by ocean container between 8 and 10 percent cheaper over the next few years. Then he backs it up with a number that is genuinely hard to argue with.

"Our AI for that saved us 2% of our ocean freight spend while improving transit time 20%," Petersen says. "Usually that's a trade-off. It's like you're either faster or cheaper, but not both." In an industry that has spent a century treating speed and cost as a seesaw — push one down, the other goes up — Flexport claims to have pushed both in its favor at once. It is the kind of result that, if it holds, does not merely improve a company. It bends a curve that touches the price of nearly every physical object that crosses an ocean.

Freight Email Forwarding

To understand why AI matters so much to Flexport, you first have to understand how unglamorous the work actually is. Flexport is, in Petersen's telling, "a global logistics company built around a modern tech stack" — a business that helps companies move cargo from point A to point B across air, ocean, truck, and rail, ideally on time, in full, and cheaper than the alternative. But strip away the tech-company sheen and the day-to-day reality is startlingly analog. "I've often joked it should be called freight email forwarding," Petersen says, only half-kidding. So much of freight forwarding is a human taking documents and sending them on: parsing a contract, confirming an address, booking a container, chasing a delivery slot by phone.

That is precisely the terrain where large language models thrive. Most contracts in logistics, Petersen explains, arrive as "giant Excel files, thousands of rows, and a dozen tabs." You cannot simply feed that to a model and get clean structured data back — but you can have AI write the parsers that ingest it. Multiply that across every email, every phone call, every piece of routine correspondence, and you begin to see the scope. "It's an endless list," he says, "and we feel like we don't even know all the things that they can do. It's still pretty new."

Scale Economies, Shared

The mechanism behind Petersen's confidence has a name he clearly relishes: "scale economies shared." The logic is Costco's, a business he professes to love "even though I don't shop there." The bigger you get, the cheaper you get; you pass those savings to the customer, who then does more volume with you, which makes you bigger and cheaper still. "Logistics is a very scale-driven industry," he says. "The bigger you get, the cheaper you get." Automation, in this framing, is simply another form of scale — a way to keep driving the price down without waiting to physically get larger.

And the labor math is stark. In the freight-forwarding layer of logistics, Petersen estimates, roughly 10% of the cost is human labor. Automate enough of that and the transportation cost of international freight could fall by around 8%, maybe 9%, over the next few years. Flexport measured that it had automated 20% of the work at the start of the year; it expects to finish the year at 50%, with a goal of 80% next year. The ceiling once looked like 80%. "Now we feel like it's probably closer to 90 to 95," he says — and rising as the models improve.

The Incumbent's Secret Weapon

Petersen occupies an unusual seat in the AI conversation: he is, as the hosts note, one of the first guests to run a company at scale that was founded before the AI wave. He has been "personally obsessed" since ChatGPT launched in November 2022, and he is candid about the internal struggle to make an established organization move. "Come on, guys, we can't be this boomer company," he recalls telling his team. "Everybody needs to be using this."

His favorite framing is competitive folklore: "We're the only large logistics company founded since the web browser." But he knows the disruption cuts both ways. "I know there's a kid in the next YC batch who's saying, 'Hey, we're the only freight forwarder founded since ChatGPT in November 2022.' And he's got a point." So why does Petersen believe the incumbent wins? Three advantages, he argues. First, the sheer scale of proprietary data. Second, the domain expertise to know which problems are actually worth solving — some of which "are maybe a feature, but not a company." And third, distribution: when a large company ships a great AI product, "the next day it can be used by thousands of companies," while a startup has to beg for data, earn trust, and win the customer one at a time. On top of that, Flexport controls its own code, while many rivals treat technology as an IT service they buy — some still running the business on Windows remote desktop.

Where the Ideas Come From

Some of Flexport's most important AI moves did not descend from the executive suite. They bubbled up from hackathons, which the company now runs "religiously" twice a year, with 50 or 60 teams and a genuine free-for-all mandate to build anything. The shift has been dramatic: 18 months ago, maybe four or five projects touched large language models. In the last two hackathons, Petersen estimates, roughly 90% were LLM-based. More striking still, these hacks stop being throwaway toys and become real product lines.

That has forced a reckoning in Petersen himself. He describes a personal arc from "way too much manager mode" — the romantic belief that you hire smart people and get out of their way — through his own "Chesky moment of founder mode," becoming far more top-down and directive. But the hackathons pulled him back toward the middle. "I never would have come up with that idea in a million years," he admits. Now he is rethinking the calendar itself, hoping to run hackathons before the roadmap-budgeting exercise so a great idea can actually get funded when it appears.

Flexport has also institutionalized the impulse. It built a 90-day AI boot camp for non-engineers: with a manager's sign-off, an employee gets one day a week to learn "vibe coding" in tools like Cursor and Streamlit and to build their own lightweight apps and workflow automations. The pitch from the program's creator was audacious — return people "10 times more productive than their peers." Petersen is honest that the metrics don't yet show a tenfold jump, but the philosophy is elegant. Rather than the old fantasy of hiring engineers under a "bait and switch" only to have them automate their way out of a freight job and revolt, you take the domain experts already doing the work and let them automate themselves. It started quietly in Flexport's Amsterdam office, "without me knowing about it," and is now going global.

The Agent That Picks Up the Phone

The concrete deployments are where Petersen's vision stops being abstract. The most customer-loved feature began, fittingly, as a hackathon project: natural-language analytics. Instead of writing SQL or building dashboards, a Flexport customer simply types a question and gets back graphs, charts, and tables. That matters because about 25% of account-management time had been spent helping people generate reports.

Then there is the container-planning model — the source of that headline 2%-cheaper, 20%-faster result. On a given week, roughly 2,000 containers get canceled by customers when a factory runs late. Software can do what no human army could: sweep through the system ten times a day, notice a lost container, and grab a slot meant to depart a week later, pulling the cargo forward. Layer a solver on top to hunt the cheapest contract, and you get both savings and speed at once.

But the moment that draws audible surprise from the hosts is the voice agent. Confirming a warehouse delivery appointment used to be a costly phone call nobody wanted to make every time — so it often didn't happen, and trucks got lost chasing bad addresses. Now, if Flexport hasn't delivered to a site in the last three months, an LLM agent handles it by email and, if necessary, by voice. "It'll actually call the warehouse and be like, 'Hey, can you confirm that 2:00 p.m. tomorrow is an okay time to deliver this?'" Petersen says. It is, he notes, a case where AI isn't just replacing work — sometimes "the work would have been too expensive, so you just didn't do the work."

Elsewhere, AI reads sentiment in the messages customers send through the Flexport platform, automatically escalating an unhappy client to a manager. "There's a lot of emotion in logistics," Petersen says. "It's your stuff. Your business is on the line." And in customs brokerage — where human brokers benchmark to about a 2% error rate — an AI "spell checker" catches the classic slip of an Australia code where the goods were actually made in Austria.

Money Wants to Spend Itself

Asked what advice he'd give founders now weighing enormous funding rounds in a heating AI market, Petersen first waves off the premise — "every company is super unique, so don't listen to advice on the podcast" — then offers some of the sharpest advice in the conversation. Capital, he says, is a beautiful thing, and only three things ultimately matter: whether you control your company, whether you have a job, and your price per share. Dilution doesn't hurt you if the price per share keeps rising.

What he underappreciated, and now takes "very, very seriously," is a subtler danger. "The part I underappreciated is the degree to which money just wants to spend itself," he says. Every company has problems, and a full bank account tempts you to solve each one by hiring someone — until you're bloated, slow, and culturally dependent on headcount instead of ingenuity. His prescription, which he says only one founder has ever actually followed: raise the big round if it's an up round, then impose a 90-day hiring freeze to signal that "the money's not going to solve our problems. We're going to solve our problems."

Coins, Prophets, and the Axial Age

It is when the conversation turns to the human stakes of automation that Petersen becomes most himself. He is unbothered by the fear that AI will eliminate jobs, because he thinks it misreads both the purpose of companies and the nature of people. "The role of companies is not to employ people," he says. "It's to deliver goods and services. Whoever employs the least number of people will have the lowest cost and win." As for how humans will earn a living, he waves it away with an appeal to bottomless appetite: "There's an infinite desire inside the human soul that can never be satisfied. We need more stuff. We got to have more."

Then he reaches for history. Around 500 BC, in what scholars call the Axial Age, coins spread across the world and quietly rewired human trust. Transactions between two people became impersonal — you no longer had to know your neighbor, check the ledger, or ask whether a stranger owed you money. "Just, here, take this thing." That convenience, Petersen argues, corroded the tight web of local relationships and bred a kind of social breakdown. And at the exact same moment, four great teachers emerged across the globe: Buddha, Laozi, Confucius, and Socrates — each, in his telling, a response to a civilization asking how to behave in a strange new world. The internet did something similar at scale, he suggests, and AI is another such rupture we have "not even begun" to reconcile spiritually or philosophically. Half in jest, he nominates YC's Gary Tan as a candidate for the next Socrates.

The hosts push back gently, and Petersen agrees the panic is overstated. Worrying about what humans will do, he says, is like standing 5,800 years ago and asking what everyone will do "when modern agriculture comes," or fretting over the monks whose transcription jobs the printing press erased. There will be real implications for society, morality, and how people relate to one another — but human nature, he insists, doesn't change much. Give people more stuff and they don't quit; they want more.

Humans in the Loop

One theme circulating among young founders, the hosts note, is that AI systems will still need humans in the loop — sometimes by government mandate. In fintech, an AI can't unilaterally approve loans; in customs brokerage, Petersen confirms, "we have to have a human that's approving the transaction before we clear customs." He sketches a provocative future in which businesses run on hyper-intelligent AI at their core, constantly optimizing across every system of record, with humans attached as "liability sinks" — the mandated approvers — and as the relationship-holders who still, sometimes, choose a vendor because of who took them to the nicest steakhouse.

But he draws a firm line. As long as there are humans in the system, they will want to relate to other humans. "We're pretty far from humans preferring to work with AI than to work with other humans," he says. And crucially, if you ever reach a world of AI serving AI, "you don't need to learn that much from the record of human history — there's no more humans involved in the loop. I don't have a lot to say about that."

Flexport in 2035

If Petersen were starting Flexport today, he says, he hopes it wouldn't be that different — because the thing Flexport got right, where a parade of pure-tech companies failed, was refusing to be a pure-tech company. "We're willing to pick up the phone and solve problems with humans, drive down to the port." He recounts a new customer needing a crane on a truck for an unusual unload — not something Flexport typically does — and simply telling his team to drive there and follow the truck to make sure it went well. "No tool use for cranes," one host quips. The mistake others make, Petersen says, is assuming that "if there's no API, I can't do it. If my agent is unable to do this task, I guess the task can't be done." You should not, he warns, try to automate that last tail of gnarly, physical, human problems.

The destination is a company where you "ship anything anywhere by any means, any mode, in any quantity, and do it all via code" — APIs or voice — so that brands never have to think about logistics at all, freeing them to do the only thing that matters: make something people want. Today the gap is geographic as much as technological. Flexport shipped cargo to and from 147 countries last year but has employees in just 22. Its roadmap targets covering 95% of all container trade with its own people by 2028, and being "everywhere that's legal" by 2035. It is a vision far larger than the customs-broker pitch of his YC Demo Day, and Petersen sounds almost giddy about it: "If you told 25-year-old me that your job this year is we've got to launch Flexport in Indonesia, Australia, Japan, the Philippines, Turkey, Poland, and France, I'd be like, 'I get to go to all those places and talk to the locals?'"

It is a fun moment, he says, and a hard one — "but fun kind of challenging. No better kind." For a man trying to automate the ocean, that may be exactly the point: the machines will handle the endless, invisible, unglamorous middle, and the humans will keep doing the thing humans have always done — wanting more, and figuring out how to move it.