Tariffs didn’t bring manufacturing jobs back to the US

Today, I’m talking with Evan Smith, who is cofounder and CEO of Altana, a company that develops software tools to manage big, messy supply chain networks around the world.

We last had Evan on in early 2025 to talk about how Trump’s first few waves of tariffs were starting to affect global trade and what patterns Altana was seeing in all that macro-level data about shipping and trade. It was a very alarming and also very illuminating chat; Evan and I got into the existential weeds of international relations and economics almost immediately. It was a great conversation.

A year and a half later, and everything is somehow even more chaotic than it was the last time Evan and I talked — and that’s true both about trade and in the business world. And, as you’ll hear, we jumped right back into the weeds yet again.

Developments in agentic AI have changed what Altana can do, but like every software company, they’ve also changed how Altana can do it. You’ll hear Evan describe how AI is changing how they make software for a very demanding set of clients that manage billions in international trade — which is causing some big debates about how a software company even runs in 2026.

And of course, what they do also needs to change more or less daily. Trade policy is changing rapidly around the world but likely nowhere faster than here in the US. Between the Trump administration’s whack-a-mole tariff policies and war in Iran, it’s very hard to keep up with the current rules.

That’s where Altana comes in. All that data coming from all around the world gives Evan a really high-level view of what’s actually happening. All those tariffs, for example, were supposed to bring manufacturing jobs back to the United States. So I asked Evan directly: Is that actually happening? Spoiler alert — you won’t be surprised to hear him say “no.” And while it’s clear the world still relies on global supply chains, the pressure placed on economic chokepoints like the Strait of Hormuz is going to get higher than ever.

There’s a lot going on in this one — as you can tell, Evan and I really enjoy talking to each other.

Okay: Altana CEO Evan Smith. Here we go.

This interview has been lightly edited for length and clarity.

Evan Smith, you’re the co-founder and CEO of Altana. Welcome back to Decoder.

It’s good to be here again. Thanks for having me.

I was looking over our last interview from about a year and a half ago, and boy, we got straight into the existential weeds of what it means to run the global economy and what was changing. People can go back and watch, read, or listen to that interview. It’s very good. I really enjoyed it.

The thing that strikes me about the past year and a half is that some of this stuff was theoretical the last time you were on the show. You founded Altana on the idea that globalization as we knew it — what you call, “globalization 1.0” — was going to come to an end, it was going to get more complicated, and the world needed software to solve it. The rubber has hit the road. It has gotten more complicated. It’s getting more complicated by the day. Even as we speak this week, it is getting more complicated. Even over the last year and a half it feels like it has gone from being a bet to being something very practical now. Is that how you see it?

Completely. One of our investors described Altana as, “an index bet on global dislocation.”

That’s a lot. Can I just ask you maybe a philosophical, emotional question about that? I get a lot of feedback from our audience. You have CEOs on the show and they talk about making bets in this way, and on the ground, it’s chaos. So, I get an index bet on dislocation. That sounds great on CNBC, right?

I understand exactly how that plays to that audience. How do you feel about that approach to the world, especially as a software company that kind of just makes a map for people, and how it plays out for regular people? I think regular people right now are really struggling and they’re really confused. They’re very anxious.

Well, trade touches virtually every part of or most of our lives. So, our mission is to fix globalization, but we’re not retreating from globalization. We’re leaning into it. But we believe that globalization needs to be fixed. It needs to be more trusted, more secure, more fair. That’s a recognition that trade will and should continue across borders. That’s the animating ethos in the company. We work with our customers who in most respects are on the front lines of all this stuff. We work with eight of the 10 biggest logistics providers. We work with the government agencies that have to scramble to enforce these laws. Then, we work with importers and their supply chains to comply with them. And the stakes are huge. Do medical devices get to the hospital? What’s the price of fuel and food?

So, I couldn’t agree with you more that the dislocation ultimately has the most profound consequences for everyday people, but that’s the point, right? Trade is the lifeblood of growth. This is how we feed ourselves, sustain ourselves, and protect ourselves. So, how do you have both at the same time? How do you have more enforcement and more fracturing as geopolitics play out, but have more trade, more growth, and more economic security? I think Altana is the answer to that question.

The reason I’m asking is that I’m worried that there’s too many layers of abstraction — not with you but with everyone. Building solutions for our current world. You just acquired a company called Cervo AI. Literally as we’re speaking. The press release is dated today.

I’ve had a smile on my face all day, but yeah, it just hit the wire.

It makes an AI platform to solve customs brokerage, which was not a problem in an earlier version of globalization.

The trade was flowing and products were moving across borders. Now, all the walls are up and everyone has to do more paperwork. And here’s Altana like, “We see an opportunity. We’re buying an AI platform that literally fills that paperwork at borders.” I look at that and I’m like, “Well, that was very smart.” It feels like the right answer for one problem. It also, to me, feels like the dumbest problem. We’re now having robots fill out forms presumably for other robots to read just to get back to the trade we had before.

I get it. A couple layers of abstraction away, this is the smartest bet. Then, down on the ground, I’m like, “So, it’s an AI that fills out forms that we didn’t have to fill out before.” The only rational response to that is another AI system on the other side reading the forms. What are we doing here?

Just to be cheeky with you for a moment, we do have our own AI agents and AI systems on one side of the border talking to our AI systems and agents on the other side of the border.

I don’t think it’s just founder bullshit to say that the future is some version of agentic orchestration of the trade network. That’s a sensible statement whether or not there’s policy volatility out there or not. The technological revolution that enables that is moving at pace, so we’re surfing that wave, we’re applying these technologies, and in some cases, we’re inventing a few. But in order to have global commerce, to have strawberries in the wintertime in the US, you have to somehow solve this border complexity problem. Your notion of if we shot ourselves in the foot is a fair question. You and I talked about this on the last episode, but I think no matter what, the US, Europe, the West, and even China in its own way, is reckoning with new geopolitical and economic security calculus. So, in a world of disequilibrium — which we’re in — where you don’t have a hyperpower that can police the entire world, all of the maritime shipping lanes, and the rest of it, there is going to be a whole lot more weight placed on economic and national security. We’re seeing the weaponization of choke points through the supply chain and global logistics network.

So, it’s not theoretical. It’s happening. We’re all reading the same news. Back to the border and to customs, the question again is, “How do we have more security and more enforcement at the same time we have vibrant trade and growth?” With all that complexity, I see no other way than to solve it with artificial intelligence.

I want to come back to that. I did it with you again, we immediately got into the weeds. Like I said, you were on the show about a year and a half ago. If people want to take a deep dive into Altana, I think that episode is a really good one, and I recommend it. It was a lot of fun. But just briefly here, describe what Altana is and what it does for people who need to get caught up.

Altana connects the public and private sectors into a network for managing trade. It’s a shared map of the world at the underpinning. So, everyone sees the same supply chain network and the same set of facts. We actually have a collaboration layer on top of that with AI helping all the parties. That now connects government agencies like US Customs and Border Protection (CBP) with eight of the world’s 10 biggest logistics providers that actually move all these goods around the world. Then, there’s Fortune 1,000 companies and their suppliers.

Think about it as hooking into a shared view of the world, like Google Maps for the supply chain, and then these parties transacting with each other to manage trade. This is where the Cervo AI acquisition we were just speaking about comes in. Agents are increasingly working alongside humans in those interactions within an organization and between organizations.

I want to dive into that specifically because I’m really curious about how the data works. I know you have a very unique data model for how people work with Altana. I know you have an entire thesis there.

But I want to ask the Decoder questions in a bit of a lightning round so we can get there quickly. Last time we spoke, you had about 240 people. I’m assuming you’re growing. How big is Altana now and how are you structured?

I think we’re actually benefiting from some of the AI we’re purveying. So, the company’s just under 300 people. We might have just broken 300 with this acquisition. How are we structured? We are trying to run more of the company through an extended leadership team these days. So, we’re covering an enormous amount of ground. I think we work with nine governments around the world. We have three product lines that serve different government users. We have this logistics vertical. We work with enterprises the world over and their suppliers. It’s incredibly broad and it’s incredibly complex.

The point of saying all that is because within a 300 person company, the communication and coordination challenges are just massive. One of the things I’m trying to solve is how to get more of the leaders in the company seeing the whole playing field, having the same debates, and synthesizing the information at whole company level and not just their function.

So, it’s messy and people complain that, “Oh, we had this two hour meeting. It was very expensive because we weren’t talking about my thing.” But my instinct is that it’s the least bad way to run the company these days as we gain scale and grow into the ambition of the business where we’re covering so much ground.

That I think connects directly to my other Decoder question. The last time you were on the show, you actually told me your process for making decisions was rapidly evolving. You said you had moved from a consensus model in areas where you had a lot of expertise to being really decisive, particularly on product and product management, and that you were going to be more opinionated and faster there while leaving the other stuff behind. That’s kind of what you’re describing at scale now. Is that still the framework? Has it evolved more? How are you making decisions?

That’s definitely stayed true. What’s happened since then is that as of March of last year, we have injected folks into the company that have really, really deep domain expertise in some of the areas we’re working in, such as trade, customs compliance, logistics, and procurement. With the advent of agentic design and agentic coding, one of the things we’re really leaning into is how to get to the right product judgments faster since it took founder-level conviction and synthesis to make some of these big bets before.

That’s still true, but what’s been really cool over the last six months is that as we’ve brought in these really deep domain experts, we can take traditional product design, product management, and discovery function, and pair it with deep domain expertise to get to the right product judgment so much faster. Then, the execution of that workflow or the build-out of the modeling steps is also much faster in this new way of coding and developing. I would just say that we have to move fast because the world’s changing really fast. That is true, and it’s probably going to stay true. What’s different from a year and a half ago is that you now have this agentic product development life cycle that just gives you so much more velocity and the ability to get to the right place faster.

Talk about that a little bit more. I’ve talked to CEOs of bigger companies that are much more stable and much older, and there’s an idea that there’s some kind of grand shake-up between what a product manager does, what a designer does, and what an engineer does, and that everyone’s going to get the same skills and maybe that classic trio doesn’t have to exist anymore. The bigger companies have got to manage that pretty carefully. They’re architected around those roles in very specific ways.

I think it’s fair to say Altana is still a startup, right? You’re still operating it a bit like a startup.

It sounds like you don’t have any of that baggage. You can just start over and say, “My domain experts are now going to be product designers in whatever way that that works.” Are the tools good enough for you to do that as a software CEO, or do you still have the other roles backstopping everything?

Yeah, it’s getting pretty psychedelic. It’s all blending together. My favorite has been watching some of our engineers lean in to design. You’re seeing it in all directions, but there’s definitely a convergence of those functions where a designer can ship code and a product manager can design and prototype. A project manager can ship code and just handle tickets themselves. Engineers can get in on the design game.

I’d be lying if I said we had the end state fully realized, but this is an active conversation that we’re having out loud with all the stakeholders in the company. These roles are converging, what are the new ways of working, and how are you going to expand your skillset and push into this new horizon? I think almost everybody’s been excited to be along for that ride and try to surf the wave, but it does threaten certain egos. If you’ve come up as a close-to-the-metal engineer, this is threatening. If you’ve come up as a really principled user experience researcher and product designer, this can be threatening. So, just like anything else, change is hard and you got to manage it through a cultural transformation of the organization.

Altana is not that old. How much transformation are we talking about here?

Plenty. We’re coming up on eight years now. I would say the difference is that we’ve got a team that’s pretty mature for a tech company because of the scale, ambition, and complexity of what we’ve taken on. We tend to hire people later in careers. Obviously, there’s a spectrum, but especially on the engineering side especially, we’ve got some really, really seasoned people. The same is true in commercial. So, it’s not just about how old is the organization. It’s about how long have folks built mastery in these domains.

I was going to ask you about that. Your clients are not lax customers. They have opinions. As you said, the stakes are very high. You’re not building productivity software, right? How do you manage the stakes of building that software against the pull to re-architect how the software is built? I know you could manage the stakes with the old way. Those old ways are proven and tested. You have a lot of people at your company. It sounds like they know those ways. You’re far less tested with the pull to build companies around new ways of making software.

You’re probably putting your finger on the raw nerve between me and our engineering team. Look, it’s getting incredibly easy to prototype and get an MVP out there. It’s still hard to get highly functional, highly performant, at-scale software deployed with service-level agreements (SLA) that include 99.999 percent uptime for workloads at some of the most important organizations in the world. So, it’s a tension. We certainly haven’t solved it. We’re starting to chip away at the edges of some of these things. I’m sure you’ve heard of and had conversations around harnesses?

So, putting real scaffolding around the code base and knowing how these systems interact with each other, follow rules, and test themselves. That’s getting better and better, and it’s getting more degrees of freedom to ship quickly, to run those tests, and make sure things aren’t breaking. But it’s not a panacea, and we’re not in the promised land with infinite coding and infinite roadmaps as much as the hype would say otherwise.

Boy, does the hype say otherwise. There’s the global economy you and I are going to talk about, and there’s the global economy right next to it that is entirely built on some of that hype.

You have a bunch of talented people who know what they’re doing. Then, there’s your customers, and I’m guessing your customers are very conscious of their margins, right? I’m guessing they push you on cost because you’re just an added cost to getting the goods to the end user and selling it for whatever money you’re going to sell it for.

How are you managing your token costs? The idea that you’re going to burn a lot of tokens inside these harnesses to get to some good outcome might increase your costs more than hiring people. This is the dynamic that we’ve heard about so much. I’m guessing your customers are not like, “Wait, you cost more money because you’re burning more tokens inside of your harness?”

Well, let’s have the conversation first about internal costs. In my observation, every tech company and everybody in the tech industry is having the same AI conversation at the same time. It’s funny, as soon as I heard it on a podcast, we started internally saying, “You know, some of these frontier model costs per token are just getting crazy, and what are we going to do to get observability and controls? Should we flip the permissioning model where you have to get permissioned into Claude Fable and not the other way around?” Everyone’s having that conversation at the same time. But it’s like $20, $50 bucks, 1 million tokens. So, the economics are kind of forcing the conversation.

That being said, we haven’t really seen token spend impact our bottom line in a material way. Sure, has it gone up on the margin? Yes, but I think we’re shipping faster, we’re being more productive, so it’s a no-brainer ROI. It hasn’t even gotten to my consciousness except for, like I said, a few weeks ago with Fable.

In terms of flowing through cost to customers, we have different sets of products. Some of them are scaling on the dimension of the products under management — the physical goods under management. So, if you’re on the Altana network, you’re managing your products, their parts, and the value chain networks associated with them. You can share those product passports with your customers, your freight forwarder, or your regulator. So, that’s the network we built. The more products or physical goods that you manage on the Altana network, the more we charge you. That’s one dimension of it. That doesn’t necessarily scale with AI compute. It does in a little way, but not from our pricing standpoint.

What we’re now reckoning with with this Cervo acquisition and some of our agentic workflows that we built ourselves is having agents do work. Agents can run up a lot of compute. So, that is a gross margin question that you have to manage carefully. Then, you get into pricing dynamics with customers where it’s like, “Well, what’s the value of the work?” Charging for tokens is kind of silly. We’re going towards charging for outcomes and for units of work where there’s some alignment around the value of work. To bring it back to the acquisition we’re announcing, in customs brokerage, you have an existing pricing model where these logistics providers are charging the businesses whose goods they move on a per customs declaration basis. It’s anywhere from $20 on the very low end to $250 on the high end for more complex customs entries. You do that day after day after day, shipment after shipment after shipment, and it kind of runs the cash register.

With agentic AI being able to do most of that work now, it puts you in some really interesting places, both as Altana and as a global logistics provider to experiment with pricing. Does it make sense to have a $200 per entry pricing model, or can we kind of shift the value creation left and say, “Hey global freight forwarder, we are going to help you manage your supply chain network to be more compliant and to be more resilient. We’re going to help orchestrate your supply chain and we’re going to stop billing you transactionally to clear goods across the border?”

It strikes me with this acquisition, in particular, the governments and people of the world are putting up a lot of trade barriers for a lot of reasons. You have a view of what you might call the path of least resistance. How do I get this object from here to there with the fewest amount of forms? It’s funny because the forms represent one kind of friction, and you can just point the gun at it. You just point the agent at it and overcome the friction.

Do you find that the AI systems are getting more efficient at that? In general, my criticism of AI is that it’s not actually learning. If you fill out the form one time and then you run the agent again, in many ways, it’s starting over from scratch. You’re going to burn all the same tokens again. Maybe the inference will get more efficient, but the compute task itself does not get optimized, even though it’s been repeated so many times. Is efficiency there for you?

Yeah, there is. In the world of LLMs predicting the next token, you’re right. In a world of agents that follow standard operating procedures (SOPs), I think you’re wrong. I’ll narrow it specifically to where I know what I’m talking about, which is this trade, customers, and supply chain world. So, with the Cervo platform — now part of Altana — you can say as a logistics provider, “I’m going to create standard operating procedures for every one of my customers and their data pathologies. I’ve got PDFs with these data fields with weird numerical formats. I’ve got Chinese script coming in alongside English script. Here’s how to handle these weird edge cases that would break earlier systems.”

You can feed the machine a document to distill rules, like standard operating procedures, and to handle that just like a human would follow those rules. They also have the user experience where, as the broker interacts with customs entries and data coming from the supply chain on a customer-by-customer basis, they use a workflow where, as the user modifies something here or there, the platform auto-suggests an update to that standard operating procedure. Then, you can go and key in an SOP from scratch.

So, it’s learning SOPs, it’s codifying SOPs, and it follows SOPs. Is there la theoretical way that those SOPs could be broken or where there’s a fail case? Surely. But in general, it’s amazing. This is why agentic AI in the general logistics space is having its moment. Every fifth company in Y Combinator is doing something around logistics automation where it’s like, “Oh, I can take all these unstructured and semi-structured data inputs and extract information from them, reason about them on a customer, trade lane, or outcome-specific basis, and then orchestrate a workflow.” That just wasn’t technically possible until like January.

I do love the phrase “data pathologies.” I like the idea that a company’s structured data is actually diseased in some way, and AI can solve it. It’s very good.

I hear you that I’m wrong. You can build a system around the models that gets more intelligent over time and maybe gets more efficient on a timescale. I’m just asking about literally burning tokens, like using the intelligence of the models. Is that getting more efficient or is it the deterministic systems around it that help the users update these standard operating procedures?

I’m probably the wrong person to be highly opinionated about this, but my co-founder, who is entitled to this opinion, has observed that there’s no real ROI on Fable right now.

Okay. So the more intelligent model is not helping you out. I’m curious about this because Altana, as a software platform, solves a lot of very bureaucratic challenges as trade barriers go up.

There’s another set of challenges I want to ask you about, which are essentially military challenges. We’ll come to that. But it seems like it’s a bit of a cat and mouse game between the amount of bureaucratic trade barriers we can put up and the amount of software we can deploy at those barriers to overcome them. At some point, the cost curve might bend and the bureaucracy might win.

Well, I’ll tell you our product strategy and our overall thesis on flipping the script. Right now, every time a shipment arrives at the border, it’s as though it’s the first time the government has ever seen that shipment of goods.

This is an abstract government or our government?

Every government in the world.

What I’m saying is true. So the customs process has been built up for millennia to scrutinize a shipment, levy the goods, and then let it through the border. It used to be the case that when we were using cuneiform tablets and there were sailboats going up the Tigris or Euphrates River and transporting goods, somebody could look at each of the things in there, issue a levy, and that was it. As container shipping came about and the scale of goods entering at a port just exploded, along with air cargo and that velocity, it became impossible to look at all the shipments and the goods in the shipments.

So for decades, the whole concept was that we’ll do random sampling, random targeting, and we’ll let the intuition of port officers dictate what they’re going to go look at. After 9/11, that was no longer okay. A whole new technology and policy framework arose in the post-9/11 world: risk-based targeting and segmentation. The idea was we’re going to try and pick needles out of the haystack using analytics. Those were pretty crappy with the systems that existed at the time. I’m not going to name names, but one major government I know intimately well has a 0.5 percent targeting hit rate in its customs targeting system. These are rules-based systems where if a container of frozen squid comes in, search it for cocaine. That’s a real example.

They found cocaine one time in a box of frozen squid, and every single shipment of frozen squid thereafter has been opened. Guess how many times they’ve found cocaine in the box?

I’m going to guess it’s zero.

So, that’s been the state of the art for the last 20 years. Now, our point of view is that with AI, two things are possible. One is that you can continuously monitor and screen everything. You don’t have to selectively target. Two is that as these trade policies are increasingly focused on the goods and their provenance — I’m sure we’ll get to that conversation, but network shape and supply chain network is the dimension of trade analysis — it becomes imperative to have a sense of product identity instead of just going transaction by transaction at the border where it’s Groundhog Day and you’ve never seen this [good] before. I would have to reason about whether this is the right customs code or the right duties and whether the Food and Drug Administration got the permit to release these pharmaceuticals or whatever. We ought to just have a library of trusted goods in the same way that we have a register of trusted travelers.

And that’s what we’re building. So, an Altana Product Passport for the US government is a library of known, continuously monitored, and vetted goods and their production pathways where in one shot you can look at the compliance attributes, their national security dimensions, the safety dimensions, the tariff dimensions of it, and have that all continuously reviewed and monitored. The benefit to the importer is they do it once.

I remember going through the airport in the early days of trusted traveler programs and thinking that there’s no way in hell I was going to give the government any information about my travel, my personal life, any of that. I don’t have to. Then, you see the benefit of the security and travel facilitation. Well, I think I’ll make that trade. That’s where we’re enabling the private sector to engage with government agencies around the world in a new way. So, we’re doing this now at scale in the United States. We’re doing it at scale in the context of defense procurement. We’re going to be doing this in a pilot context in Europe and I think the UK in the coming months.

There’s a lot to unpack there. I’m curious about the Product Passports. Again, this strikes me as a solution to an administrative or political problem. The Trump administration particularly wants to move manufacturing out of China and reduce this dependency on China. A lot of companies, big companies, want to move their supply chains to get them out of those risky parts of the world. Maybe they want to clean up the environmental or labor impacts of their supply chains. You provide all that to them. We can vet this, this product is going to show up. Like I said at the top of the conversation, the rubber has really hit the road on all those policy goals.

Are you seeing manufacturing move out of China into the United States? You’ve got these passports. Are you seeing the origins of those passports change in meaningful ways? Do those passports change in meaningful ways?

Yes, although probably not exactly in line with the policy objectives of the administration. Without getting into political commentary, which I’ll mostly skip, I’ll give you some points of view as a reader of the news and of the data, and as an observer of the supply chain network itself through our own platform. One question is if the United States is reducing its dependency on China as a source of manufacturing. I think the answer, controversially, is no.

So, what’s happened? We see this in our data. We see this in the macro statistics as well. Since these trade barriers, in particular the Liberation Day tariffs and everything since, you’ve seen a proportionate decrease in Chinese exports to the United States with a proportionate increase in US imports from third-party countries like Vietnam, Mexico, Canada, and Malaysia. We can see through our own platform where those input flows of goods are coming from, and the answer is China. So basically, we’ve just rerouted Chinese goods through third-party countries where they’re undergoing some amount of transformation (or not) and are on their way to the same end market destination at a higher cost. We can say that numerically.

Now, that’s part of the dynamic behind the renegotiation of the United States-Mexico-Canada Agreement (USMCA), for example. I’m sure we’ll talk about that, but this notion of supply chain traceability becoming the buzzword in all of these industrial policy and trade policy initiatives is at the core of it. What’s the actual provenance of the goods? What’s the multi-tier value chain network of the goods? We have to know that if we’re going to have policies that try to reshape the network itself. So, that’s one dimension of the China question.

I think another one is, have we created more jobs? Have we created more manufacturing in America? There, I think it’s kind of a mixed picture. The Purchasing Managers Index (PMI), which is an index of manufacturing output in the United States, went down after the tariffs. So, we were declining as an economy following the big wave of tariffs last year in terms of our manufacturing output. In the last five or six months, that number’s actually trended above 50 percent on a scale of one to 100, but it’s above the neutral level. It’s now actually increasing and it has been increasing. I don’t know that this is true, but one way you might observe that is in the initial shock. You have US manufacturers that are importers of stuff in order to make more stuff. If the input availability of those things is highly disrupted, then so are their supply chains and their output. I know anecdotally from our own customer base, that was absolutely the case.

So, the first impact of the tariffs was actually negative for US manufacturing. What might be playing out that we should be open-minded about is that as the network adapts to the new tariff environment, you’re seeing a general tilting of the scales in favor of US manufacturing and therefore an increase in the PMI and the rest of it. What might also be playing out is that the AI data center thing is peanut buttering over everything.

Then, just the last thing on manufacturing is, are we or are we not creating a bunch of manufacturing jobs in America? Again, no political statement here. Just kind of observing the data. On a net basis, the answer is no. So, manufacturing jobs continue to decline. How do you square this increase in manufacturing output that we just talked about with the decline in manufacturing jobs? The answer there must be industrial automation. I think it was always kind of naive to imagine that we’d go back to the assembly lines of the 1920s in America. Is it important to have economic security for a country? Yes. Is it important to have the manufacturing capacity to make very important things on your own soil or with trusted allies? Yes, it is. Are those efforts going to lead to an explosion in blue-collar, well-paying jobs? That is a question that has not been answered.

I do think we can leave aside the political question of who in the administration sold a very naive vision of what would happen after these tariffs and who didn’t. But on a policy basis, the goal of the tariffs was to increase costs on goods made out of the country. So, we would manufacture them here and it would become economically viable to manufacture things here.

You’re saying basically that has not happened. And if it does happen, those factories will be automated anyway. We had the CEO of Siemens on the show, which builds factories for people. He said straight up, “Yeah, I’m going to put robots in Kansas. That’s the way we’re going to do it if we do it at all.” Are you seeing that kind of explosion of interest? Are you seeing that kind of response? There’s USMCA news literally today that we will come to, and maybe that will shake everything up again. But it’s been a year and a half since Liberation Day, since that whole conversation. Whether or not that actually happened, I think you’re confident in assessing just based on the numbers.

I think at the macro level, it seems clear that manufacturing is generally going up after going down at first. Manufacturing jobs are not, and industrial automation would explain that. I have a front-row seat to a few different industries, just given the work we do, and I’d say there’s evidence of the industrial policy working. I’ll give you one: drones, or autonomy more generally. So, the Chinese control roughly 90 percent of the relevant value chain components for autonomous systems. Even if you have a US startup that’s saying, “Hey, we’re making everything here in America,” no, you’re not. You’re assembling component parts to make an air or land-based drone, and you’re saying it’s made in America.

We had Adam Bry from Skydio on the show just a few weeks ago. This was legitimately the conversation he and I had. It was, “where does it come from?”

We are helping that industry with this question and with that problem statement, but it’s not going to happen overnight. What I will say is that a couple policies actually came together in a coordinated way to create a real US market not just for the end platforms but the components themselves. So, on the one hand, you had a bunch of FCC restrictions and import restrictions generally that said none of the logic or system controls can come from China. They have to be US made and have to have US parts, or at least non-Chinese in the fine print. What that did was effectively turn off the supply chain for these platforms and make everything more expensive.

To pair alongside that, what they did was they created a large source of demand and revenue. In the Department of War, there’s now this program called Drone Dominance, which is billions of dollars of committed capital. So, there’s a strong market signal. Actually, autonomous is playing a role in certifying the value chain. How do you have component parts that are trusted and where there’s real provenance? It’s pretty wild. You’re seeing a little Shenzhen forum in El Segundo with all these defense tech bros who are building autonomous systems. That’s kind of what it takes. You have to shape the market sufficiently to achieve your policy objective. Some of it is like putting up barriers. Some of it is putting up outright bans in this case. Some of it is subsidies and some of it is directed investment.

That’s the other piece of it. Over the last 15 months, the administration has gotten incredibly aggressive on not just throwing up barriers but actually making offensive moves. So, it’s grants, it’s equity investments, it’s loans with warrants, it’s choosing winners in the private sector and then just plowing capital and procurement into them. We’re in the process of creating a national stockpile of critical minerals and magnets.

Methods aside, is that the right kind of policy outcome to aim at? Yeah, I think so. We learned over the last year and a half that you don’t have economic security if you don’t control critical minerals. So, you’ve got to do something.

Let me ask you about that. Again, last time you were on the show, there was the idea that a bunch of warehouses would pop up in Canada and Mexico to import almost complete products from China so we could get them over the border inside the USMCA. I would say it was like a hack. You knew that was going to happen. You could see it beginning to happen.

We could see it happening.

The other thing along that same line is that in September, the United States ended the de minimis exemption for low value imports. So, there used to be a threshold where you didn’t have to pay duties at $800. There was no customs declaration owed. In order to exploit that, you had those Mexican and Canadian distribution centers where containers full of stuff were coming in, but then being drop-shipped one by one across the border to US consumers. So, the same has been true in Europe and the UK, and they too just ended their de minimis exemptions. So, it’s not just the United States adding all this complexity and friction to businesses and end consumers, this happening everywhere.

Just this week, President Trump announced new tariffs on Canada. Maybe because he’s mad about wildfire smoke, maybe because he’s mad that Spain won the World Cup. It’s unclear exactly what happened to provoke a new round of tariffs. US Trade Representative Jamieson Greer just did an interview with The New York Times and he’s like, “We’re going to run it back on tariffs. We got smacked down at the Supreme Court, but we take that as an invitation to find a new legal mechanism to do tariffs the way the administration wants, the way that President Trump wants.”

It is unclear to me what legal foundation the new tariffs against Canada are resting on and whether they will be upheld. Just based on what we have been talking about here — politics and policy outcomes aside — it feels like the tariffs broadly did not do the things the administration wanted. There are some things that worked better, particularly plowing capital into markets and picking winners and losers, which is not a thing that people generally love when the United States government does it but it has worked in some cases. Yet, we’re going to do tariffs again.

I’m curious for your view on how trade with Canada will change now that we’ve announced some tariffs that may or may not exist, that may or may not get rebated over time. More particularly, with your view of the network, how do you see the network adapting to them?

Here’s my take on Canada.

The Canadians have not come to the table on renegotiating the trading relationship with the United States. I know that from some of the behind-the-scenes stuff I’m exposed to but also because it’s reported. Whereas, Mexico and the United States have been very constructively working on what will become the new rules of the USMCA, in particular supply chain traceability, which is kind of the linchpin of the United States’ new framework. I’ve got to steer clear of the politics, but the way I see this latest round of Canadian tariffs… so this is Section 338, which I admittedly hadn’t heard of until a couple of days ago.

With 48 hours to consider it, I actually think 338 has a really strong legal foundation for the specific complaint in this case. So, 338 gives the president the authority to implement counteractive measures when a trading partner is selectively biasing or penalizing US trade relative to other countries. And the Canadians did just that over the last year and a half on autos, on dairy products, and on alcohol. There’s no US alcohol on the same shelves where there’s European alcohol. There’s plenty of legal authority for the claims themselves.

Where it’s going to get interesting is the remedy. Section 338 says that the remedies have to be proportionate to the damages. So, were the tariffs, the 50 percent across this peanut butter of Canadian imports, proportionate to the three or four industries that were called out? Probably not, butt how long is it going to take to adjudicate that? It’s going to be a long time. In the meantime, everyone’s working to put something new in place for the USMCA since the US opted out of the agreement and now there’s kind of a shock clock to get some new rules. So, I kind of see the whole thing in the context of them trying to force the Canadians to the table around the USMCA renegotiation.

I can see that in the abstract of, we’re doing geopolitics and having a good time. We’re all playing Risk together. But from a policy perspective, Liberation Day happened. You’re saying it may or may not have had the numeric effects we wanted it to have. Some other policies had good effects. Then, we ended up paying a lot of rebates because the legal foundation wasn’t solid. Donald Trump and John Roberts are going to be in a fight that’ll be very interesting to legal academics 100 years from now. But right now on the ground, we’re about to have that fight again, and we might rebate all those tariffs again. That happened with the Liberation Day tariffs at higher rates.

It certainly could play out that way.

You run the software to manage all this administrative and bureaucratic complexity. How do you see the tariff rebates playing out inside your system? Does that tell us anything about how companies and markets might react to this new round of tariffs?

We’re certainly helping our customers with the rebates. We can calculate duties at any point during the time of the then-applicable law and do that over a bunch of complexity, like you said. I think this is the new normal. Again, it’s not just the United States. The Chinese have put in place the mirror image of the United States from an export licensing standpoint. They’re compelling the private sector to get certificates,permits, and permission from the Chinese to take critical minerals, battery technologies, a whole set of things, and they can trace those all the way through the end uses and end markets through a big global network. The Europeans are doing the same thing. The Europeans just gave themselves the legal authority to do a whole lot more from the trade barrier and trade enforcement standpoint as a unified customs authority.

So, I don’t see the world going any other way than with unilateral behavior where geoeconomics is part and parcel with geopolitics and it’s getting expressed mostly as these trade policies but as economic security more broadly.

I think that brings us, sort of inexorably, to the next turn, which is the warfare turn. We can see it playing out today in the Strait of Hormuz, right? The United States launch an attack on Iran, Iran shuts down the Strait. We’ve basically brought global shipping in that area to a dead halt. We’re now maybe in a much longer war about who will control the strait and what those tolls will be.

Altana has a role to play there, too. You can watch goods move around there, and you can reroute the goods. Put that into context for us. That is the next turn. It feels like we can play a lot of economic games and a lot of policy games, but when we start firing bullets, everything changes in very dramatic ways that are hard to deescalate. Describe what you’re seeing there and what the pathways out of it might be.

I think the lesson that everyone’s learning in Hormuz is that economic choke points can and will be wielded against adversaries, and that they can and will lead to live, physical warfare outcomes. It’s no longer theoretical. It’s no longer, “Hey, I’m sanctioning these banks,” which is kind of what you and I grew up with. It’s any of these economic and supply chain choke points, especially if they’re asymmetric — meaning it’s more painful for you when I turn you off than it is for me to lose you. A great example of this involves rare earths and critical minerals in China. The dollars that US end markets generate for Chinese critical minerals providers are a rounding error. Nobody cares. But when they turn those off, they can shut down the United States’ ability to make a weapon. So, it’s highly asymmetric, and those can and will be exploited. And those can and will turn kinetic. I think that’s the world we’re entering.

Literally, last time on the show, you and I talked about John Mearsheimer, who spent his entire career predicting that we would go to war with China. He was my professor at the University of Chicago in the late ‘90s.

Do you think that’s inevitable? The promise of Altana is, “We’re going to fix globalization. We’re going to make trade easier.” And the promise of trade, particularly global trade, is you’ll have so many economic interdependencies that war will be unprofitable.

I think the extent to which you can de-lever to those asymmetric choke points and economic weapons impacts whether there will be peace. For example, if the US and the EU can make semiconductors and have supply chain and economic security with advanced electronics in the event of a Chinese blockade in Taiwan, then there’s a much lower probability of war. I think you can kind of apply that same argument across all these choke points.

So, yeah, I think supply chain resilience and redundancy, having the network capability to see and manage through that, and simulating and designing more resilient systems is the best path short of controlling the switch yourself as a president. That’s the best path to creating a world that’s more likely to be peaceful.

The Strait of Hormuz, in particular, is fascinating because it is oil. We can see what’s happening to energy prices around the country and world as that conflict waxes and wanes in ways. Well, it only appears to be waxing. There are moments of waning. It’s like one of those iPhone charts. Cumulative sales always go up. That’s what appears to be happening there.

Do you see that energy prices are affected by the number of ships day to day in the data, or is the system responding to the shock?

Both are true. What’s so cool about Altana is that you get so many layers beyond the headline. How is the shock to the helium supply out of the Gulf cascading through semiconductor value chains and pushing on costs there? We can actually give you numerical answers to that, and we can show you the dependencies and the pathways.

I think in general, we — and when I say “we,” it’s the pundits, the policymakers, and the people who play Risk — tend to both over and underestimate the substitutability of goods. That’s been super interesting because it’s like, “Oh gee whiz, these things are going to turn off.” Then, there are lots of examples where they were much more substitutable than folks knew and/or the trade routes could lengthen and you add 40 days to a voyage but supply chains still flow. In other cases, you’re learning about choke points that are not at all substitutable. Those Achilles’ heels reveal themselves.

Altana has kind of a bingo board of these things, but what I’m really interested in is, how can we get to a world where you’re not discovering those the hard way? How do you get to a world where you can detect them, you can simulate them, and you can build scenarios that actually shape either policy or capital allocation? We have a whole product, R&D area, and the way to think about it is that we’re enabling policymakers and firms to simulate these things, build those plans, and do the long-range network architecture.

I want to end here because I know you have a lot of thoughts about network architecture and federating data, which is maybe the most pure Decoder bait of all.

Altana is a system. It works best if everyone participates in the system. You obviously spend a lot of time convincing people to participate in the system. That means they have to offer you some data. They obviously have to get some data out of the system in return to be valuable. You’ve talked about AI quite a bit in this conversation. There’s a lot of concern about AI just hoovering up all of the data and using it for whatever purposes the labs have, eating businesses like yours or maybe not even protecting that data at a base level. That is the architecture of your product, right? You gave a keynote presentation about federating data.

I want you to end by just talking about that. There was a version of the world you and I grew up in which multipolar, multi-party organizations across the world would just show up and share data and make everything easy. They were geared towards cooperation. You have been describing unipolar actions over and over again in this conversation, and Altana is supposed to make that a little bit easier, right? It’s a system that will connect a bunch of individual rational actors, and the data has to connect in a way that everyone is comfortable with. That seems very hard. Can you just describe how you are getting everyone to participate in this? Because it feels like a software solution to a politics problem.

It is a software solution. The other way I’d describe it is it’s a market making solution. So, a lot of our calories are spent from government affairs or sales standpoint bringing parties together that have aligned interests in these novel ways of working to solve a network shape problem.

So at the highest level, Altana is a network. You join a network and you benefit from what the network provides. So that’s the connectivity to other parties in the network and the intelligence and visibility that’s revealed in the network. As you connect to the network, you are contributing to that network.

What we are not doing is sharing everybody’s data with each other, and that was the point. There’s no world… there certainly wasn’t back when we started the company, and it’s becoming even more true now. You’re not going to reach this nirvana data commons where everyone’s just pooling data in one place.

So, how do you have shared visibility? How do you have interoperability across global commerce in a world where sovereignty is ever increasingly important? Our answer to that is federation. You have to put the software and the data in control of the actors that require it. It’s non-negotiable, especially at the scale of working with a customs agency or a defense agency.

So, we are architected in this kind of hub-and-spoke model where the spokes are the customers with very large sensitive data sets that need us behind a firewall. They might even need us even in front of a firewall but with kill switches, corporate governance, and so on. What we’re doing is we’re bringing the platform to their data. We have very specific rights to the learnings. As we connect to the data and process it, we have the rights to build out the supply chain network view. So we’re not taking pricing. We’re not taking a detailed bill of materials. The supply chain connections themselves are revealed from the data as we connect this supply chain graph.

Then, the other bucket of rights are around learning with pre-trained models. What we can’t do is memorize our customer’s data. What we can do is take a pre-trained model and, for example, we have systems that detect risky shipments in global trade or we predict the valuation or customs classification of goods. Those models are pre-trained. They’re learning across the network, and as we learn from any given customer’s data, those model weights and configurations can be lifted off for the benefit of the network. That is the quid pro quo. You give a little and you get a lot. Like I said, the go-to-market and government affairs is increasingly about putting together these parties at the highest level.

I’ll tell you a couple vignettes. We’re very proud of our platform in the UK. We provide the underlying infrastructure for what they call the Global Supply Chains Intelligence Program. It’s this whole UK government supply chain and economic security control tower. I think there are 10 agencies that are a part of this. They’re pouring their data into the platform. We bring the global picture. We bring the software and the models. They’re reaching these supply chain resilience and economic security policy outcomes that I think are world leading, and they’re doing this on this intelligence platform that underpins it.

We’ve been putting them in the room with senior folks in the United States, Australia, and the Netherlands, and we’re helping them learn from each other on the art of the possible. We’re helping them imagine ways of potentially coordinating policy responses across these agencies and collaborating on some of the actual interventions in the supply chain. Like, what if we all collaborated on traceability in the critical minerals context?

I don’t think it’s unnatural for these parties to join the platform. I do think it’s novel for them to imagine some of these new ways of working, and that’s kind of the friction in the sale. I’m proud of the progress we’re making. It’s like any other network business. The more mass it builds, the more inevitable and more valuable it becomes.

Let’s end by just zooming out a little bit. We’ve described a world in which we inevitably move towards more war. We’re just going to go to war against China because there are choke points and people are going to squeeze them. Then, there’s your vision. We’re going to imagine new ways of working together in which maybe we’re not part of big NGO organizations that manage trade globally, but everyone’s acting in their rational interest. We have a view of the network that makes it easier to imagine new ways of working with all kinds of supply chain traceability.

Tell me where we are in the journey of those two things. End there. Make it not abstract. Bring it all the way down to people on the ground who are feeling anxious about the state of the world today. Which way is it going? Which way do people want?

I think we’re going toward a world that is more economically secure and, therefore, more physically secure. By knowing the provenance of goods; by designing supply chain networks and actually cutting through the fog to make these decisions proactively; by helping the public and private sectors engage with each other constructively and collaboratively on these dimensions; the arc of history is moving toward more economic security. Therefore, there will be less motivation to go to outright war in the event that those economic dependencies are weaponized.

Will there be flashpoints? Of course there will. Is there going to be a world of friction? Of course there will. I think at the end of the day, everyone wants to be safe. They want their kids to be safe. They want their wealth to grow. They want to spend time with people they care about and not have to sweat this stuff. In a world that’s transitioning from one equilibrium to the next, I’m optimistic about the direction of travel here, even though it’s going to be messy.

Evan, it is always a pleasure to have you on the show. Thank you so much for being on Decoder.

_{Questions or comments? Hit us up at decoder@theverge.com. We really do read every email!}