The AI industry largely remains focused on building better models. Every few months, a new benchmark appears, context windows expand, reasoning capabilities improve, and developers gain access to more capable systems. The progress has been substantial, but many of the problems that prevent AI agents from operating autonomously have little to do with intelligence itself. Instead, the challenge is infrastructure.
An AI agent may understand a request, create a plan, and figure out the necessary steps, but it still faces a problem that humans solved decades ago. The agent needs to know what services exist, which one can perform the task, how to access it, and how to complete the transaction.
That gap between reasoning and execution has become one of the most significant bottlenecks facing agentic AI, and it is the problem Orthogonal is attempting to solve following its newly announced $4.3 million seed round led by Pantera Capital and backed by Y Combinator, Pioneer Fund, Decasonic, Blast, Outbound, Surreal by waPremise, and other strategic investors.
The company, founded by former leaders from Coinbase, Vercel, Google, and Amazon Robotics, is building what it describes as a discovery, orchestration, and payment layer for AI agents. While much of the AI market continues pursuing better models and more capable assistants, Orthogonal is focused on the systems agents require once they leave the chat window and begin interacting with the broader internet.
The distinction may prove increasingly important as companies move from experimental AI deployments toward software that can complete real work.
For years, software integrations depended upon humans acting as intermediaries. A person searches for information, chooses a service, authorizes a payment, and verifies the result. AI agents promise to remove portions of that process, but doing so requires infrastructure that was never designed for autonomous agents.
Before Google, humans could still navigate, integrate, and transact across the web, but it was slow and fragmented. AI agents face the same challenge today, and Orthogonal provides the infrastructure to make it seamless.
Current agents often perform well in controlled environments but struggle when they encounter unfamiliar services or workflows. When new capabilities are required, they may hallucinate, fail outright, or require human intervention to continue. The underlying models may understand what needs to happen while lacking any reliable mechanism to make it happen.
Orthogonal approaches this problem as an internet infrastructure challenge rather than a model challenge.
The company’s platform currently supports more than 40 APIs spanning data, sales, and business services, allowing agents to discover capabilities in real time, orchestrate multiple services, and execute payments within the same workflow. The company describes the approach as similar to how model routers direct requests to different large language models, except the routing occurs between agents and the services necessary to complete tasks.
That distinction reflects a broader shift occurring throughout the AI ecosystem. The first phase of generative AI largely centered on models themselves. The second phase focused on applications built on top of those models. Increasingly, attention is turning toward the infrastructure required for agents to operate reliably across the internet.
McKinsey has projected that agentic commerce may represent $3 to $5 trillion in economic activity by 2030, but those forecasts depend upon agents becoming capable of executing tasks independently. Booking travel, conducting research, enriching business data, purchasing services, or coordinating workflows all require systems that extend beyond language generation.
The internet itself was built primarily for humans. Search engines assume a person will review results. Websites assume a person will navigate interfaces. Payment systems assume a person will authorize transactions. APIs often assume developers already know which services they need. Autonomous agents challenge those assumptions because they require infrastructure designed specifically for software acting on behalf of users.
This is where Orthogonal’s founders appear to see an opportunity. The team’s experience across payments, infrastructure, cloud computing, and developer platforms reflects the increasingly interdisciplinary nature of AI infrastructure. Building autonomous systems requires expertise that extends beyond machine learning and into payments, identity, security, orchestration, and developer tooling.
The company already supports multiple payment rails, including traditional payment systems alongside agentic payment protocols, while working to simplify payment execution for autonomous systems. That capability may ultimately become as important as discovery itself. An agent that can identify the correct service but cannot complete the transaction remains only partially autonomous.
Over the next year, Orthogonal plans to significantly expand the number of services available through its platform, moving from dozens of integrations to potentially thousands. The objective is not simply to create another developer marketplace, but to establish a common layer through which agents can discover capabilities they do not yet possess.
The broader AI industry often speaks about autonomous agents as though they are primarily an intelligence problem. Better reasoning, stronger models, and larger context windows certainly matter, but the ability to operate independently may depend just as heavily upon the infrastructure surrounding those models.
As AI systems increasingly move from answering questions to performing actions, the companies building the connective tissue between services may become as important as the companies building the models themselves.
Orthogonal’s funding announcement, therefore, says less about another AI startup entering the market and more about where the industry may be heading next. If autonomous agents eventually become meaningful participants in the digital economy, they will require their own discovery systems, payment rails, and orchestration layers. The intelligence may attract the headlines, but the infrastructure may determine whether the agent economy actually functions.
