F2 Turns Private-Credit Memory Into AI Infrastructure

F2 is building an agentic operating system for private credit, commercial banking, and private-markets investment teams. Founded in 2025 and led by CEO Don Muir and CTO Emre Kazdagli, the New York company connects the work between a new data room and an investment decision: screening, diligence, financial modeling, underwriting, investment-committee materials, and portfolio monitoring.

The sharper idea is not that artificial intelligence can read a PDF. Plenty of software can perform that party trick. F2 is trying to preserve the part of a credit firm that usually disappears into shared drives, departed employees, and yesterday's model: the institution's own judgment.

That proposition has become more concrete in 2026. F2 disclosed $24M in total equity funding in June, introduced its Adam deal-team agent, expanded its Institutional Knowledge product, and announced a Google Gemini Enterprise partnership. In September, Golub Capital added a $5M strategic investment and plans to deploy F2 across its investment workflows.

About F2

F2 emerged from technology developed inside Arc, the finance company Muir previously co-founded. Muir has described the origin in practical terms: private-markets teams kept rebuilding context from spreadsheets, data rooms, and investment memos, while the lessons from completed or declined deals remained hard to retrieve.

The company spun out in September 2025 with an oversubscribed $10M round backed by nearly 50 investors from the Arc network, including NFX, Left Lane Capital, and Y Combinator. F2's name is a nod to the Excel shortcut for editing a cell, a small detail that says more about the product than a warehouse full of AI slogans. This company was designed around the spreadsheet, not around pretending the spreadsheet vanished.

F2's customers are private-credit funds, commercial banks, and private-equity teams. The company says more than 100 funds and banks use its platform, with those clients managing more than $400B in assets. It also reported more than 15,000 deals analyzed during the year, 650% growth in monthly active users, and roughly 5x throughput across screening, underwriting, and monitoring. Those figures are company-reported, but they show the scale of the workflow F2 wants to own.

What F2 Built for Private-Market Work

F2 brings unstructured documents, Excel models, deal-room files, memos, and third-party data into a shared intelligence layer. Agents can assist with screening, financial spreading, diligence, scenario analysis, credit writing, investment-committee preparation, and monitoring.

The architecture matters because finance does not grade software on how confidently it improvises. A number in an underwriting model needs a source, a formula, and a person willing to defend it. F2 says its platform runs a native Excel engine, tracks formulas and dependencies, and uses Audit Mode to connect outputs to the underlying documents.

Its Institutional Knowledge product takes the same logic across time. Prior memos, models, risk decisions, and portfolio outcomes become a queryable library. A team evaluating a new borrower can search for comparable deals, retrieve earlier reasoning, and reuse precedent without treating an old folder structure as corporate memory.

That makes model independence commercially important. Foundation models will keep changing. A credit firm should not have to abandon its history, controls, and workflow every time the intelligence layer improves. F2's multi-model approach is designed to keep the institution's context stable while the underlying models rotate.

Why F2 Matters Right Now

Moody's expects private-credit assets under management to exceed $2T in 2026 and approach $4T by 2030. More capital means more opportunities, but it also means more documents, more structures, more monitoring, and more chances for inconsistency to sneak into the process wearing an expensive vest.

The category is moving from chatbot access to governed execution. Investment teams need systems that can work inside permissions, preserve source lineage, use firm-specific precedent, and leave a trail that survives an investment-committee question. DevCuration has tracked the same move in enterprise AI through Arcade.dev's authorization infrastructure and Ema's cross-system AI employees. F2 applies that operating problem to credit, where a polished wrong answer can become a very real asset on a balance sheet.

F2 reports SOC 2 Type I and Type II certification, GDPR compliance, zero model-provider data retention, and no training on customer data. Those controls are company-described, but they address the correct concern. Private-credit firms are not uploading restaurant menus. They are handling borrower files, proprietary models, negotiated structures, and institutional judgment.

Leadership Built From the Workflow

Muir's path runs through private equity, Stanford Graduate School of Business, and Arc, where he says the team used AI to expand debt placement from smaller transactions to billion-dollar facilities. Kazdagli led the technical work behind Arc Intelligence before becoming F2's CTO.

That history gives the founding team a useful constraint: the product has to function for people who live in Excel and are accountable for the result. The founders are not pitching a blank-page replacement for credit judgment. They are building software meant to make that judgment easier to retrieve, apply, and audit.

In June 2026, F2 introduced Adam, an agent designed to coordinate a deal workflow across specialized sub-agents while preserving assumptions, sources, and decisions. The company also named Bain Capital and Live Oak Bank among customers and announced a partnership with Google Gemini Enterprise.

The September Golub relationship adds another kind of validation. Golub is both investing and preparing to use the product. A deployment inside a manager with more than $90B of capital under management is not proof of better returns, but it is a serious environment in which to test whether institutional memory, model flexibility, and auditability survive real operating complexity.

Hiring as a Demand Signal

F2 said its June financing would support growth in engineering and enterprise deployment across New York, San Francisco, and London. The expansion makes sense for a product that cannot be dropped into a credit firm like another browser tab. Enterprise deployment requires people who can translate investment workflows, data controls, model behavior, and user habits into a system that teams will trust.

The current openings on F2's careers page may change, so the more durable signal is the kind of work behind them. F2 needs builders who can operate between financial rigor and AI infrastructure, plus deployment professionals who understand that adoption is a systems problem, not a demo problem.

That is why the hiring story belongs in a company spotlight. It indicates where demand is pulling the organization: deeper product capability, more customized enterprise work, and support for customers across multiple financial centers.

The Bigger Industry Shift

F2 is making a bet about where value will sit as foundation models become more capable. The model itself may be rented. The durable asset is the customer's governed layer of data, precedent, workflow, and decisions.

For private-credit firms, that layer is unusually valuable because judgment compounds across deals. A declined transaction, a covenant negotiation, or a risk that later surfaced in a portfolio company can improve the next underwrite only if the organization can find and apply what it learned.

F2 now has capital, public enterprise references, a strategic investor that plans to deploy the product, and a market growing quickly enough to punish manual repetition. The next proof will not come from another AI benchmark. It will come from whether investment teams can move faster while preserving the context and accountability that make a credit decision defensible.