Ian Crosby raises $10M for Synthetic startup

Ian Crosby, the founder behind Bench and Teal, has launched Synthetic, a San Francisco-based startup aiming to build an agentic bookkeeping system, and has raised $10 million USD in seed funding led by Khosla Ventures, BetaKit reports. BetaKit names angel backers including Shopify CEO Tobi Lütke and Basis Set Ventures, and reports Crosby is leading a five-person team that has been working on Synthetic since last year. BetaKit quotes Crosby directly: "I don't know if this is possible yet," and reports he described the product as an AI agent that connects to banking, payroll, billing systems, and inboxes, then asks humans targeted clarifying questions to resolve gaps. BetaKit says the system remains in a "crash test phase" and still makes mistakes, and it records Crosby noting accounting demands near-100 percent accuracy for customer trust.

Agentic bookkeeping, as described by Crosby and BetaKit, implies several technical challenges common across the industry. Companies building similar systems typically need robust connectors to financial systems, deterministic reconciliation logic, strong entity extraction and classification, reliable human-in-the-loop query generation, and auditable reconciliation trails. Model hallucination and error rates create particular risk for bookkeeping use cases because errors map directly to financial statements and regulatory exposure. Teams pursuing agentic workflows often combine ML components with deterministic rules, transaction-level verification, and explicit provenance to satisfy auditors and customers.

Founders with prior exits or notable startups often attract investor confidence for high-risk projects; BetaKit documents that Khosla Ventures led the round and several high-profile angels participated. The combination of experienced founders, seed capital, and a small engineering team fits a typical early-stage pattern for startups attempting to productize complex ML systems. For practitioners, the crucial vector is proving operational reliability under real accounting workflows rather than solely improving model metrics in isolation.

For observers and potential customers, track evidence of enterprise pilots, third-party auditing or SOC2-style controls, release of connectors to major payroll and banking platforms, accuracy or reconciliation metrics on real data, and noise-handling or escalation patterns for cases the agent cannot resolve. Editorial analysis: Metrics that tie model outputs to verifiable ledger reconciliations will be the most persuasive signals of product-market fit in this category.