Overview
Zomma is a San Francisco startup founded in 2026 that describes itself as an "AI BPO" for financial services run by computer-use agents: AI agents with their own computers that operate the same screens human analysts use — custodian portals, portfolio accounting systems, desktop applications — with no APIs, no integrations, and no IT projects. The bet is a deliberate wedge: target back-office work where mistakes are expensive (compliance, reconciliation, risk), and convert compliance fear from an objection into a buying reason by making human approval gates and audit trails the core of the product rather than a limitation. The company pairs forward-deployed enterprise engagements with a waitlisted developer API ("Give your agent a computer"), hedging between services revenue and platform scale. Its YC profile lists it in the Summer 2026 batch with a five-person team and an "Active" status label, and PitchBook lists its stage as "Generating Revenue" — though no named customer, revenue figure, or post-Demo Day funding round is disclosed, so the traction story rests almost entirely on company statements. ycombinator.com founderland.ai pitchbook.com
Founding and team
Zomma was founded in 2026 and went through Y Combinator's Summer 2026 batch, with YC partner Brad Flora working with the team; the batch's Demo Day was scheduled for September 10. CEO Jihyun Kim's stated background is unusually finance-heavy for an agent startup: she says she launched a $1B hedge fund as founding team and Head of Product, scaled Asia's largest wealth tech (now $10B+ AUM), is a second-time founder, and is ex-Goldman Sachs, with degrees from Cornell and NYU Abu Dhabi. CTO Cashel Fitzgerald holds a Cornell CS master's, built aerial object detection systems at Modern Intelligence, and engineered backend systems at 8am supporting $300M+ annual revenue and $20B+ yearly transactions; his stated specialty is training vision-language models for computer use. CPO Varun Arumugam became co-founder in January 2026 per his LinkedIn, after a founding ML engineer role at Bronco AI building agents for chip design verification and a software engineering stint at Apple on verification infrastructure. The pattern is notable: a finance operator paired with two engineers whose prior work was verification and detection — the exact skills a trust-first agent product demands. No first-party account of how the founders met or why they chose this problem exists in the record, and Kim's prior company is unnamed. ycombinator.com linkedin.com founderland.ai
Product and positioning
Zomma's positioning is blunt: "Zomma starts where mistakes are expensive: compliance, reconciliation and risk work in financial services." Its agents read what's on screen, reason through tasks using what the company describes as frontier language models, execute actions via simulated mouse and keyboard, and verify results by reading the output again. The target workflows are the unglamorous core of financial operations: reconciliation between custodians and portfolio accounting systems, 1099 reconciliation, NAV tie-outs, corporate action processing, regulatory filing preparation, client onboarding across CRM and custodian portals, periodic KYC reviews, and compliance logging with append-only audit trails. The company says its agents navigate systems like Advent Geneva, Bloomberg Terminal, and Schwab Advisor Center, and handle fund administrator and transfer agent portal automation for hedge funds and family offices. zommalabs.com zommalabs.com ycombinator.com founderland.ai
The no-integration claim is the wedge's commercial logic. Pear VC describes the setup as training an agent on a desktop by walking it through a workflow once, after which it runs with no APIs, integrations, or IT projects required. That matters because financial back offices run on custodian portals and legacy desktop software whose owners will never build an API for a five-person startup, and where traditional RPA breaks on any interface change. Zomma's May blog post explicitly contrasted its method with traditional RPA and API integrations for insurance carrier portal data entry. The company also claims each agent is proven on a private benchmark built from the customer's own past cases before going live and again after every change — a company-stated claim with no third-party validation in the record. zommalabs.com cbinsights.com pitchbook.com founderland.ai pear.vc
Distribution runs on two tracks: forward-deployed enterprise engagements, and a developer API described as "Give your agent a computer," which per Founderland's coverage required joining a waitlist. The dual motion is the hedge in the thesis — services revenue from hands-on deployments funds the company while the API tests whether the underlying capability can scale as a platform. No pricing, contract structure, or unit economics are disclosed for either track. founderland.ai
Trust architecture as the sales argument
Zomma's most distinctive product decision is that it does not promise autonomy — it promises control. Human reviewers make every determination; agents only recommend. Per the company's YC profile, "Every action stops for human approval before anything submits." When two-factor authentication prompts appear, agents halt and wait for a person to clear them; the company states flatly, "We never bypass MFA." In a market where a fund manager's fear of an agent fat-fingering a wire is the primary purchase blocker, Zomma has made the approval gate the product. zommalabs.com founderland.ai
The company has built an unusually elaborate trust stack for a five-person team, documented across its blog and security documentation. A June post outlined an "Agentic Trust Framework" with four layers: input assurance, governed reasoning, policy-mediated action, and isolated execution, separating a policy-deciding orchestrator from ephemeral runtime instances. A July post explained why Zomma assigns SPIFFE identities to those ephemeral runtimes so every agentic action is attributable. Security documentation describes single-tenant deployments in AWS us-east-2, TLS 1.2+ in transit, AES-256 at rest, scoped credentials in a dedicated vault, 1Password integration, SSO through WorkOS, an immutable audit trail, and a commitment that customer data is not used for model training. SOC 2 Type II certification is pending — no completed audit is on record. The content marketing reinforces the same argument: a July 20 post titled "The trust gap: why 95% of fund managers use AI and only 7% run agents" (a company-cited statistic with no named source), a September 29 post on FINRA guidance for broker-dealer operations teams, and an October 4 post on testing AI agents before production. Zomma is selling the trust gap itself. founderland.ai zommalabs.com
Traction and go-to-market
Traction evidence is thin and entirely company-stated. CEO Jihyun Kim wrote in a LinkedIn post announcing the YC backing: "We already work with some of the most ambitious funds and fintechs" — but the company declined to identify specific customers, and no named customer, case study, or reference logo appears anywhere in the record. PitchBook lists the company's stage as "Generating Revenue" across two completed accelerator deals dated January 1 and March 16, 2026, but no revenue figure, ARR, contract count, or Demo Day traction metrics are disclosed. The forward-deployed enterprise model implies high-touch, likely few customers at this stage; whether any of the "ambitious funds and fintechs" convert into referenceable logos is the single most important unknown about the business. founderland.ai pitchbook.com
Economics and financing
Financing is modest and partly conflicting. CB Insights lists Zomma at $500K total raised on a convertible note, with Y Combinator as an investor; PitchBook lists Y Combinator and Vercel AI Accelerator as investors, both as accelerator/incubator type, across the two 2026 deals. The Vercel AI Accelerator participation appears only in PitchBook with no first-party confirmation, and no post-Demo Day seed round or Series A has been announced. Who pays and how much is undisclosed: the enterprise motion presumably bills per engagement and the API per usage, but no pricing exists in the record. A five-person team on roughly $500K of accelerator capital implies the company either has meaningful early revenue or will need a seed round promptly after Demo Day. cbinsights.com pitchbook.com
Competitive landscape
Zomma enters a market where both ends are crowding in. RPA incumbents UiPath, Automation Anywhere, and SS&C Blue Prism have added agentic features to existing platforms — Automation Anywhere launched pre-built "Autonomous Finance" solutions in May, and SmartStream announced "Smart Agents" for banking back offices in June, per press releases from both companies reported by Founderland. Newer entrants include Dispatch, which orchestrates account opening across Schwab, Fidelity, and Pershing; Automatic.co, which deploys back-office agents for accounts payable and reconciliation; and Skyvern (YC S23), which builds browser automation using language models and computer vision. Zomma's differentiation against all of them is the same thesis: vertical focus on financial services workflows where mistakes are expensive, plus the no-integration desktop approach versus API-based or RPA-based alternatives. No competitive win/loss data exists, and no evidence shows how Zomma's approach compares on cost or reliability to incumbents at scale — the incumbents' installed bases and the entrants' narrower wedges are both plausible counters to the Zomma read. founderland.ai
The honest competitive assessment is that Zomma's moat is currently narrative, not demonstrated. The trust architecture is well-documented and unusually thoughtful for the stage, but the private-benchmark claim, the unnamed customers, and the absent pricing mean the company's advantages are self-reported. What would change the read: a named customer or case study, a completed SOC 2 Type II audit, disclosed accuracy rates on customer workflows, a post-Demo Day funding round with traction metrics, or any third-party validation of the 95%/7% trust-gap statistic. Until one of those lands, Zomma is a well-constructed bet whose evidence base is the company's own writing. zommalabs.com founderland.ai zommalabs.com
Outcome and evolution
No acquisition, pivot, or shutdown is on record, and no post-Demo Day financing has been announced. What the record establishes is labels, not verified status: Zomma's YC profile lists its status as "Active" and PitchBook lists its stage as "Generating Revenue," both database designations rather than audited facts. The company's dated public output — a September 29, 2026 post on FINRA guidance and an October 4, 2026 post on testing agents before production — shows continued marketing to operations leaders through that October. Beyond these labels and publications, the company's present operating status is not established by the cited evidence. The story so far is a coherent wedge — trust-gated computer-use agents for expensive-mistake financial work — with the outcome still unwritten. ycombinator.com pitchbook.com zommalabs.com
