Fragment raises from FirstMark to build programmable money infrastructure for AI agents

Fragment is building what founder Thomas Neckel calls "programmable money" — a financial system purpose-built for AI agents to automate the back-office work of commerce. The company has raised from BoxGroup (pre-seed), Stripe (seed), and most recently FirstMark, which led a new round announced alongside this conversation. The round amount was not disclosed.

Neckel's path to Fragment runs through a prior company, Cove, which handled automated fund administration for private investments including real estate and private equity deals. Administering those deals with a two-person team — including for Party Round — exposed the core problem Fragment is now built around.

“We wanna automate commerce. And we're gonna do that by outsourcing the work of finance to agents... Commerce is a set of promises — instructions for who owes what, to whom, when. And money is the system that settles them. You take the rules from all these contracts, you put it next to the balances that those rules need to update, and now you have programmable money.”

The core argument

Neckel's framing is that commerce is fundamentally a set of promises: instructions for who owes what, to whom, and when. In B2B specifically, those instructions live in contracts, PDFs, invoicing terms, SLAs, refund rules, and billing logic — distributed across separate systems with no unified view. Humans currently carry context across those systems. As transaction volume grows, that human-in-the-middle dependency becomes an economic bottleneck.

Agents, in Neckel's telling, can't readily resolve that fragmentation today — not because they can't navigate systems, but because the rules of money need to be deterministic. An agent disbursing an insurance claim, for example, needs to reconcile a claim balance from one system, payment history from another, and accounting entries from a third. Without a unified layer holding those rules, the agent lacks the context to act correctly.

Fragment's answer is to extract the rules from contracts, place them alongside the balances those rules govern, and let agents execute against that unified structure. The intelligence stays with the model; Fragment constrains it from acting on incomplete or conflicting state.

Where it sits in the stack

Neckel is explicit that Fragment is not competing with Stripe or Plaid on payment rails or connectivity. It operates one layer up — a ledger API and system of record on which fintech companies build their financial products. Ramp and WAP are named as customers already running on the infrastructure.

The agent-specific buildout is recent. Over the last six months, Neckel says, those customers began using agents to run back-office workflows on top of Fragment's ledger in ways that pushed the system into new territory. That usage pattern is what shifted Fragment's product roadmap toward agents explicitly.

The company is based in New York and actively hiring go-to-market roles.