San Francisco-based startup Sapiom has emerged from stealth with $15 million in seed funding to solve one of the most persistent bottlenecks in the agentic AI revolution: how to let software pay for software. The round was led by Accel, with participation from heavyweights including Okta Ventures, Menlo Ventures, Anthropic, and Coinbase Ventures. The company is building a dedicated financial infrastructure layer designed to handle the complex authentication and micro-payments required when AI agents need to purchase their own tools, data, and compute resources.
As AI development shifts from simple chatbots to autonomous agents capable of executing complex workflows, a new friction point has emerged. While tools like Lovable and other "vibe coding" platforms allow non-technical users to generate applications via plain language, these AI-generated apps often hit a wall when they need to connect to the outside world. Currently, every time an agent needs to send an SMS via Twilio, spin up a server on AWS, or access a paid data stream, a human must manually intervene to set up an account, enter credit card details, and manage API keys.
According to reports, this manual "glue" breaks the promise of autonomy. Sapiom’s thesis is that for the agentic economy to scale, software needs the ability to transact independently. The startup argues that the current financial rails—built for human decision-making and slow procurement cycles—are ill-suited for a future where agents might make thousands of micro-purchasing decisions per day.
Sapiom is not simply issuing corporate credit cards to bots. Instead, it is building a programmable financial layer that acts as a proxy between the AI agent and the service provider. The platform issues what it calls a payment and identity "envelope" for each agent. When an agent determines it needs a specific capability—such as a paid API call—Sapiom handles the delegated authorization flow.
The system generates ephemeral credentials, effectively allowing the agent to access the service without exposing a master credit card or permanent API key. Crucially, this infrastructure comes with enterprise-grade policy controls. Organizations can set strict budget limits, vendor allowlists, and spending rules, ensuring that an autonomous agent doesn’t accidentally rack up a massive bill while trying to solve a problem. This turns spending into a programmable, secure action rather than a risky open loop.
Who Is Backing the ‘Machine Economy’?
The $15 million seed round represents a significant bet by top-tier investors on the infrastructure of the "machine economy." Accel led the investment, but the strategic participation of Anthropic (a leader in LLM development) and Okta Ventures (identity management) underscores the industry’s belief that identity and payments are converging for AI.
The company is helmed by Ilan Zerbib, the former Director of Engineering for Payments at Shopify. Zerbib’s background is particularly relevant; he spent years building payment rails for millions of human merchants and is now pivoting to solve similar problems for non-human buyers. His experience in risk, reconciliation, and high-volume transaction processing is likely a key factor in attracting such a high-profile roster of investors.
Analysis
Sapiom’s launch highlights a critical maturity phase in the AI cycle. We are moving past the "model layer" wars and into the "enablement layer." Just as Stripe removed the complexity of payments for the internet economy, Sapiom aims to remove the complexity of procurement for the agent economy. The involvement of Coinbase Ventures also hints at a future where these transactions might eventually settle on blockchain rails, which are naturally suited for the high-volume, low-value micro-transactions that agents generate. If Sapiom succeeds, it could effectively become the central bank for the next trillion buyers—none of whom will be human.
What This Means
For developers and enterprises, Sapiom’s technology promises to reduce the time-to-value for AI applications by removing the administrative burden of connecting services. For the broader market, it signals that autonomous commerce is no longer science fiction but an engineering problem being actively solved. We can expect to see a surge in "agent-native" business models, where software negotiates, buys, and sells services with other software at speeds humans cannot comprehend.
Source: Original Article
