Stateful
Whether you’re a founder, investor, or simply blockchain-curious, Stateful brings you behind the curtain with the team that’s been first to nearly every milestone in the space.
Pantera Capital was the first U.S. institutional asset manager exclusively focused on blockchain technology, launching the first U.S. crypto fund when bitcoin traded at $65. Since 2013, Pantera has pioneered venture equity, early-stage tokens, and liquid digital asset strategies—backing more than 100 blockchain companies and 110 token projects globally.
Stateful
Why AI Agents Need Infrastructure Before Intelligence with Orthogonal
Use Left/Right to seek, Home/End to jump to start or end. Hold shift to jump forward or backward.
Mason Nystrom sits down with Christian Pickett and Bera Sogut, co-founders of Orthogonal, to explore why the agent economy needs infrastructure before it needs more intelligence, and how Orthogonal is building the missing layer that lets AI agents discover, call, and pay for services on their own.
The problem: an agent can decide what to do but cannot do it. Using a new service requires another integration, another API key, another contract, and another bill, each set up by a human. Orthogonal connects dozens of verified providers through one API an agent can access autonomously, paying per call in cents rather than per seat or subscription.
Key Topics:
- The missing layer: agents can reason and decide but cannot act autonomously because every new service requires a human to set up accounts, manage API keys, and handle billing across multiple providers
- Why micropayments finally make sense in an agentic world: agents follow policies, track spend automatically, and can transact at sub-cent levels that humans never would, collapsing the per-seat subscription model
- The agentic payments standards war: X402, MPP, Nano Payments, ACP, and UCP are all competing, and Orthogonal is taking a protocol-agnostic stance, supporting all of them so providers get discovered regardless of which wins
- The human internet vs the agent internet: agents will not browse websites, compare pricing pages, or fill out forms. A parallel agent-native internet with its own discovery layer, GEO over SEO, is forming right now
- Full autonomy is not the end state: human preferences need to be baked into every long-horizon task, and the agent that lives with you proactively, monitoring your calendar and deals and acting before you ask, is the next frontier
Some people think that agents will use the internet similarly to how humans do. We think that they will not.
SPEAKER_02I think it's important to always have humans in the loop. Even if it's like a long horizon like task that has multiple different steps, there's still some sense of human preferences that you would want baked in into any given task.
SPEAKER_00Welcome to Safeful, a Pantera podcast, where we talk about everything at the intersection of finance, venture capital, and tech. I'm your host, Mason Nystrom. And today we have an amazing pod diving into the intersection of AI and payments. And for that, we have our most recent Pantera portfolio investment, Christian Pickett and Bera Sogat, co-founders of Orthogonal. Welcome to the pod, guys. But before we begin, a quick disclaimer. This content is for educational and entertainment purposes only and does not constitute financial investment or legal advice. Please do your own research before making any investments. It'd be helpful to level set and just give the audience a brief rundown of what you're building at Orthogonal.
SPEAKER_02Yep. So orthogonal is where agents go to find access, orchestrate, but and also pay for capabilities they don't already have. For example, a growth agent might be looking for people information or finding emails or phone numbers or trying to enrich to get a LinkedIn profile. Right now they'd have to manage different accounts and setting them up and billing across all these different providers. But with one integration or one place to one place that you can connect to, they can easily use all these different services. You support fiat, but also many different crypto crypto and agency payments protocols from X402, MPP, and nano payments.
unknownRight.
SPEAKER_00And so there's two core parts to this. One is enabling an agent to connect to any workflow or automation endpoint that they need. And the second is kind of giving them the capability to pay for it. I guess continuing with kind of the automation component, as you onboard more of these endpoints into kind of the orthogonal marketplace or platform today, what are you typically seeing as the types of workflows that agents want to use or that developers want their agents to be able to access?
SPEAKER_01We have our main use cases: are companies wanting to find leads, wanting to find companies that they want to sell their products to. We have many providers that provide these kinds of data. So people come to us wanting to find generate a list of 100 companies to sell their product to, and then they want to find their contact information. So emails, phone numbers, wherever they want to reach them. And we have many providers that do this as well. And also the titles, who they want to reach at these companies. Oh, I want to sell to the CTO of this company. Can you find me the CTO? The decision makers. So we have different providers ranging from cheap to expensive. The agent does not have to create an account and does not have to manage all these different API keys.
SPEAKER_00Why can't an agent just use Stripe or a credit card? Why do we need this kind of new type of payment infrastructure to enable agents to be able to access these services?
SPEAKER_02Yeah, like credit cards solve the payments problem, but not necessarily the discovery, authentication, or account management problem that's there. For example, for the GTM agent use case, if an agent wants to access a couple of different providers, such as like Apollo, people data, core signal, link up, those are multiple different subscriptions, potential contracts, credentials, API keys, billing that the agents will have to set up, manage. And then also Stripe has isn't able to handle micropayments as well. So provo service providers have to set minimums to make the unit economics work. So we from the beginning have been supporting paper call or paper request sent-based transactions, which makes more sense for agents.
SPEAKER_00Right. This idea of micropayments historically has kind of failed as a business model in previous tech environments, in part because, like you mentioned, a lot of human usage has typically been like per seat or per subscription, and humans haven't been as willing to pay for a little bit of something, or rather, enterprises don't want people to pay for a little bit of something and instead want to charge on that per seat model. I'm curious, like, how do you see micropayments and potentially this model making more sense in an agentic first world?
SPEAKER_02Yeah. And you stated the problem that micropayments have had. I'd also say that users, people, the everyday person don't want to be able to track all these sense that's that they're paying for. Or also, like with micropayments, often people are comparing it to paying for an article, which it's usually free to view articles at times. So it's also the mental burden that people have and also the comparison to free. But agents themselves are able to follow specific policies. So a person can just say, hey, I just want to spend $10 for this task or $10 a day. And then the agents can go ahead and call multiple different services and pay these set subcent transactions and also keep track of how much it's spending. So agents aren't necessarily having to offload that burden of tracking these really small transactions.
SPEAKER_00Right. And is this are these types of micropayments a behavior that you're seeing happen in real time and where rather than an agent having to buy a subscription, they're paying cents for each API call to these services?
SPEAKER_02Yeah, that's what we're seeing right now. We're seeing, as Bear said, people are adding our MCP to their cloud code or to their agents, or they're using their wallets using MPP or X402. And they're giving the request for, okay, I'm looking for specific data, be that social media or people data, and they're making those requests, getting, doing those small transactions and getting that information. So that's what we're actively seeing right now on our platform.
SPEAKER_00Yeah, I'm sure a lot of people, when they think of an agent conducting any type of payment for them, they think, oh my God, what if the agent hallucinates and it pays $1,000 instead of a cent for something or it buys too many of an object? Bear, I'm I'm curious, like, how do you think about preventing an agent from overspending or buying the wrong service or getting tricked by the malicious provider? What is the policy engines that you think have to exist? And how does orthogonal think about that problem?
SPEAKER_01We're actively working on this. We're now right now, users can set daily spend limits, how much they want their orthogonal account to be able to spend on a daily basis, and also allow lists on which APIs you can access. So our endpoints range usually from a cent to a dollar. So yeah, if they don't want to use the APIs that are costing like a dollar, they might they can disable those ones. So then yeah, the agents can only use what they're approved, but we're actively working on making this better. Maybe like prompt limits per transaction limits, more like category restrictions. We also will be tracking like provider reliability, latency, and accuracy across all calls as well. And yeah, the and also we're not a free-for-all marketplace. We value quality over quantity. We vet every provider, they need to be legitimate and already have real users. So yeah.
SPEAKER_00Right. And so you're taking a white label approach to the services that come onto the platform. And as you think about building this payments layer for a variety of other applications or developers to use, one of the things that we've seen emerge is of a variety of payments companies try to build their own standards, right? Like Coinbase is building X402, Stripe has MPP. I'm interested to hear your thoughts on really the various competing standards that exist and how you think the standardization war plays out into the future. Do we settle on one or two standards? Do we have 16 standards that like you end up just integrating for everyone in a seamless experience? What is what does this end state look like?
SPEAKER_02Yeah, I could say this. Yeah. I would say that we started by seeing X402, which is which I love, and it was basically repurposing the HTTP standard and incubated by Coinbase. And I think for like when we when Bear and I first saw this, we're like, okay, it makes sense. Agents should be paying using microtransactions using stable coins. But now we've seen Tempo release MPP machine payments protocol. We've seen Circle release nano payments. We've also seen OpenAI and Stripe work on ACP, that's a genetic payments protocol using payments intense at the core. But also Google and Shopify working on UCP, setting up a standard for merchants to make their products discoverable. And honestly, it's not clear which protocol is gonna win. There hasn't been a lot of adoption of the ACP just yet or UCP, but we're taking the stance of being a bit agnostic to whichever protocol wins and supporting all of them. That's why right now we support X402, which we started with. We support MPP, we support nanopayments. We're also supporting multiple different blockchains as well. So we see this more as okay, the standards are coming. Let's figure the market's gonna decide which one wins based on where the transaction volume is going. But we ultimately believe that service providers are going to want and need to have their services, digital services, accessible by agents. And they're gonna want to be discoverable across all these different standards too. So embedded into orthogonal means you're discoverable across all these different protocols.
SPEAKER_00Right. And on the flip side of that is kind of the standard standardization of the payment method. A lot of these obviously standards allow you to pay in whether it's credit, whether it's stable coins, or from like an agent's wallet. How do you envision this evolution of payments is gonna happen? Do you think that eventually what will happen is most agents just have access to a credit card and they'll pay via credit card? Are they gonna pay via credit card and then that'll stable settle in stable coins? Are they just gonna pay in stable coins directly? What do you how do you think this plays out?
SPEAKER_02I personally want everything to go through stable coins and an open ledger, but the on-ramps are just not there yet. And right now there's many different companies trying to give agents credit cards or debit cards, and almost all checkout flows support credit cards and debit cards and not necessarily wallets. So we're kind of in a stage where it's much easier for an agent to use their credit cards to purchase something. But I believe it's a bit more elegant to have these microtransactions going over an open blockchain in stable coins. There is that in between where it's credit card, but ultimately set settles on stable coin, blockchain-based infrastructure underneath. But I believe that should be fully abstracted from the day-to-day consumer. So so I think we're yet to see. And for me, I ultimately just want the on-ramps to get improved overall.
SPEAKER_00I totally agree. I think there needs to be improvements on the on-ramps to make it easier. Maybe zooming out for a second, I assume a lot of people are like in a novice stage of using AI today, and that's typically through chat or Claude cowork. Do you personally see a world in which Claude and maybe chat are more dominant, or do you think we'll see a lot of verticalization around the types of transactions that people will have in the world, like having a personal agent for your travel, having a different agent for your Amazon shopping cart and groceries? Like, how do you think about the expansion of agenda commerce more broadly?
SPEAKER_01In my opinion, yeah, I guess Sam Altman and also the creator of OpenClaw have quoted that most apps will become APIs in the future and mess the messaging apps will still be there. But if our agent is able to do all of the other things we're doing, like ordering food or ordering Ubers and things like that, if it's able to do that, why not? So I do think, yeah, most things, transactions will happen through an agent. And I would think that it would make sense if there's a general agent that we talk to that might maybe talk to other more specific agents. But I think eventually there will be just one agent that we talk to.
SPEAKER_00How does that shape your viewpoints on where you think value accrues in this adjunctive commerce world? You obviously have kind of the end applications, you have the payment rail infrastructure, you have agent platforms, whether it's harnesses or the actual models, and then you have new types of marketplaces like orthogonal. Obviously, orthogonal is one place that value will accrue, but like how do you think about it in terms of other parts of the stack as well?
SPEAKER_01For us, we think that yeah, agentic payments will become huge in the future, but we still need like more guardrails and agents to become smarter, and we need to like trust agents more on these purchasing decisions. So that's why we believe like starting small with small purchases like APIs. That's what we're focusing on right now. But eventually we will move on to bigger and bigger purchases. I'll let Christian like add on to that.
SPEAKER_02I would add that value accrues to where the decision is made and also where the agent goes to make the trust to make a trusted decision. Basically, around like where the agent makes a decision and also the trusted medium, be that a marketplace, be that an orchestrator, be that another platform. Because right now it's like the agent platforms are where the agents are, and that's where the user is, that's where the demand is. But then if I want to give an agent a specific task, and the task includes purchasing something, one, the human still needs to be in the loop, but also we need to figure out where the policy is. What are the white allowed listed of services or purchases that agents can make? How much can the agents spend? This can live on the agent platform, but this could also live on the on the service provider or be at an intermediary that's the manages that relationship. And I guess we are really trying to solve that problem with allowing, like setting spend limits, having allowed services, having more governance controls and policy controls. And it's ultimately that layer, that's agents decision-making layer and controls plane, is where we believe value accrues.
SPEAKER_00Yeah, makes a ton of sense. What is something that you think most people in AI believe about agents today that you think they're wrong about?
SPEAKER_02I would say that that full autonomy is the end state, honestly, and the human in the loop is temporary. I think it's important to always have humans in the loop, given what even if it's like a long horizon like task that has multiple different steps, there's still some sense of human preferences that you would want baked in into any given task. So be that okay, I give the task to to make a restaurant reservation. There's a bunch of questions underneath that. It's okay, how far is the restaurant? What type of food? What's the vibe? And that's those additional questions can be prompted or requested back to the person.
SPEAKER_01I want to add we touched on it a bit before, but I guess some people think that agents will use the internet the similarly to how humans do, but we think that they will not exactly browse websites like we do. They will not fill out like forms, compare pricing pages, and the human internet and the agent internet will look different within a couple of years.
SPEAKER_00I could imagine a world in which the agent universe actually has more economic activity happening and the human discovery layer kind of just shifts more towards the application layer rather than the horizontal search that we have today. Is that one framing of how you're thinking about those two fragmented internets?
SPEAKER_01Yeah, yeah. I would say uh the as I said, like the agents don't have to go browse websites, find services on Google, and see which ones make sense, compare which one they want to use, compare the subscriptions, like pay for a $250 subscription, for example, things like that. We think that's not very efficient for agents, and that's why we're building orthogonal. I think we want to be that efficient layer for agents to access services.
SPEAKER_00I guess one kind of random related question that is GEO, like generative engine optimization is becoming of increasing importance. How do you guys think about GEO versus SEO and the recommendation of services to agents?
SPEAKER_02We've actually been thinking about this since the beginning through working with our first API partner. They they one of the core things that they came to us for was discovery, actually. It was yes, payments, but also discovery. They wanted their APIs in front of agents, this new type of customer. And they even suggested that, oh, at some point they would be willing to pay like some kind of let's say to be a featured or to be a preferred provider. And we've started to be getting these asks from the service providers. So now there's a whole new type of economy that is like GEO, but for these specific paid services. So this is something that we're thinking about deeply. We're primarily focused on more like the agents as of now, but we see that okay, this is this has huge potential going forward. And also, we're using cloud code, like ChatGPT, all the time, and they are searching the web. And there are many different companies trying to position themselves in front of these agents, but ultimately I think the strong value around the search engine optimization is around these paid services.
SPEAKER_00Interesting. I want to move on to a quick fire round where I ask a quick question or say a quick statement, and you give me your immediate response. What does your personal AI stack look like today?
SPEAKER_02Yeah, mine is I use conductor to code, and I've also been experimenting with Whisperflow. And I've been a big fan. Haven't been I briefly play around with Fable 5, but it is huge and expensive. But sticking with Opus right now. Yeah.
SPEAKER_00How about you, Barrow?
SPEAKER_01I use Quad Code mostly. We also have an open claw agent running on our Mac Mini that lives on our Slack. So that one has its own linear account, GitHub accounts, so it's for quick changes. We're able to use that. But also it always has our company context without the session getting cleared or anything. So yeah, it's been pretty useful. And anyone in our company can use it. So yeah.
SPEAKER_00The first, I guess, employee of orthogonal was just a Mac Mini. Next question. So there's an ongoing debate that credit cards are good enough for agents and that they'll never really need blockchain wallets. Agree or disagree?
SPEAKER_02I agree in the short term that they're good because of accessibility, but in the long term, I think crypto blockchain-based payment rails is the more elegant and the correct architecture for these micropayments.
SPEAKER_00Bear conflicting thoughts? No, yeah, I agree with Christian. What does it take to get to that more simplistic architecture?
SPEAKER_02Improved on-ramps. I will continue to yell this from the tower. We need great on-ramps. And specifically around the KYC aspect, we recognize KYC is extremely important. But when someone wants to go through a checkout flow and has to enter their social insurance number, they're not going to make a purchase. You're going to lose most people. So right now, we need some way for like guest checkouts. We know Coinbase has this. We know Stripe has been experimenting with this. We just need an easy way for people to pay with the credit card and have stable coins under their hood. I know Privy's working on things in this space.
SPEAKER_01To add to that, I think we've also had a lot of users saying, like, people, when they hear crypto, they don't want to use the product. So I think crypto also needs to get a lot more trust from some people in the world. Yeah.
unknownYeah.
SPEAKER_00Regardless of, I guess, how the agent pays, like it'll still be abstracted from the end user, which is what you guys have been saying is the is an important part. And then what's something you've changed your mind on in the last year?
SPEAKER_01I guess yeah, we thought that the more autonomous agents were like years away, but now seeing like open claw and cloud code also like improving a lot more. Like that proved that they're here. And that's one thing. Also, like we thought maybe the cost of LLMs, the cost of AI is like still like a huge barrier. But yeah, right, it's getting cheaper and cheaper every day and like cheaper and cheaper over time. And we think that eventually it will be like cheap enough and this won't be a problem. Yeah. And I guess in the future, like all with all these advanced advanced investments, we think that like proactive agents will be like the next thing where like you don't like prompt, you don't have to necessarily talk to the agent, but the agent like kind of lives lives with you and like monitors your logs, your calendars, your deals, and maybe even acts before you ask, maybe asks your approval for things. So yeah.
SPEAKER_00Yeah, I like the idea of proactive agents. I think that's totally something that's that that has a ton of potential. Christian, anything else you've changed your mind on?
SPEAKER_02I think I'm I think I'm just a huge fan of open claw. And I know there's a huge hype at the beginning of the year and it's kind of died down. But having these agents just living wherever you are is the future. Is it something that I've changed my mind on? Yes, as Bear said, the timeline, I think it's here, and but it's just not I love the quote the feature's here, but it's not evenly distributed. I think that we're having that with OpenCloud, with these autonomous agents, with some of these proactive agents, and it's more of us just getting it to the world. And I think it's all of us, everyone building the space that is playing some part into that.
SPEAKER_00And then maybe lastly, what should people expect from orthogonal in the next six to 12 months? What should they get excited about?
SPEAKER_02We're gonna have more and more capabilities, more things your agents can agents can do. We're gonna introduce more interesting orchestration and workflow tool tools as well, more governance tools as well for giving your agents more control around how they're spending. Some people have said that with orthogonal your agents get superpowers because they just get access to many different things, and we're in it to give more and more superpowers to everyone's agents.
SPEAKER_00Amazing. Well, Christian Berat, thank you so much for joining.