Powering the Agentic Economy: Our Seed Investment in Sapiom

In the next phase of the internet, software won't just execute work-it will also spend money. As AI agents begin running more workflows end-to-end, they will inevitably become economic actors, procuring compute and inference, calling usage-based APIs, purchasing data, and allocating capital to services autonomously as part of their normal operation. This creates an entirely new class of spending: continuous, automated, and at a scale fundamentally far beyond anything our human-centric rails are designed for.

Currently, agents operate inside financial infrastructure built for people. Account setup, approvals, billing, and compliance all assume the presence of a human. When agents initiate spend, these systems break down. Giving agents unfettered access creates unacceptable risk, but locking them down makes them useless in production. For agents to operate safely at scale, teams need a new layer in the stack: a control plane for automated transactions that makes machine-initiated payments programmable, observable, and governed by policy.

Sapiom is building that layer, an autonomous spend API that sits between AI agents and the services they pay for. Through a single integration - 5 lines of code - teams can give agents controlled access to paid services while Sapiom abstracts away identity, wallets, risk controls, billing, and settlement. Once integrated, agents can transact safely on their own: not only writing code, but now provisioning the resources required to run it. In effect, Sapiom transforms spend into a programmable capability that software can reliably use on its own without human intervention.

These agent-driven transactions are already occurring, and the same dynamics that reshaped corporate spend are reappearing in new shapes: fragmented tooling, limited visibility, non-deterministic behavior, and major compliance risks. But this goes beyond small tweaks to a checkout flow. To co-opt a phrase Ilan shared in one of our early meetings, "agents won't make people magically buy 10x more refrigerators," but they will generate orders of magnitude more volume in agent-to-service transactions. At that scale, this stops being an implementation detail and becomes a core systems problem.

Ilan Zerbib, founder of Sapiom, spent his career operating at the frontier of payments and automation, as a co-founder of Earny, one of the earliest AI shopping assistants, and then at Shopify, where he led payments engineering. He is a hands-dirty builder through and through, and has experience standing up infrastructure for scale, growing ShopPay and building ShopCash from the ground up. Joining Ilan are engineers from Earny, Shopify, ModernTreasury, Affirm, and Intercom - a deeply technical and execution-oriented team with relevant experience across payments, LLM infrastructure, and distributed systems. I always see it as a positive when founders are able to recruit top talent from their previous stints.

Accel is leading Sapiom's series seed alongside some notable strategics such as Vercel Ventures, Okta Ventures, Anthropic, as well as angels from Shopify, OpenAI, Github, Circle, and Mercury. Sapiom joins a growing group of infrastructure companies we've partnered with, including Nebius, Tailscale, Supabase, and several yet to be announced, that are enabling the large-scale transformation driven by AI and agents.