Company Spotlight: Valon Rebuilds Mortgage Servicing

Valon is building an AI-native operating system for mortgage servicing, the decades-long work that begins after a home loan closes. The New York company connects loan data, accounting, money movement, workflows, investor reporting and compliance in one system of record so human operators and AI agents can work from the same context.

Founded in 2019, Valon took an unusual route to that product. CEO and co-founder Andrew Wang, President, COO and co-founder Linda Du, and co-founder Jonathan Hsu built and operated a licensed mortgage servicer for six years. The operation became a laboratory for the exceptions, audits and customer problems that a software demo rarely sees.

That operating history matters now because mortgage servicing is moving into an AI cycle without escaping its regulatory burden. ValonOS is already live at Carrington Mortgage Services and ServiceMac, while Newrez plans to begin a large transition in 2027. Valon has to prove that its platform can preserve correctness as it moves from a company-operated environment into some of the industry’s largest institutions.

About Valon and ValonOS

Mortgage servicing covers payment processing, escrowed taxes and insurance, defaults, foreclosures, investor remittances, borrower assistance and the record of every decision made along the way. Valon says the U.S. market touches roughly 50M homeowners and $12T–$13T in loans. Much of that work still runs across mainframes, spreadsheets, paper and tools that do not share context.

ValonOS is designed as the authoritative record beneath that work. It stores what happened on a loan and why, including the exception, approval, precedent and compliance context. Workflows and task management route the next action, while the same data supports investor reporting and money movement.

Ditto, Valon’s native AI-agent layer, operates inside those controls. Valon describes the agent as handling homeowner communications, escrow analysis, complaint research, payment reallocations and customer-specific processes. The important distinction is architectural: the agent does not sit above a disconnected pile of systems and guess at the missing record. It works inside the platform responsible for preserving that record.

The Operating Business Was the Product Lab

Valon spent its first six years running a licensed mortgage servicer on ValonOS. Operators encountered edge cases, engineers changed the data model or workflow and the operation tested the result again. That loop forced the product to absorb the details that regulated financial systems usually hide inside institutional memory.

In August 2026, Carrington Mortgage Services acquired Valon Mortgage, adding approximately 810,000 loans to Carrington’s servicing operation and adopting ValonOS as its core platform. The sale separated the operating business from Valon Technologies and let the technology company focus on licensing software to other servicers.

This history gives Valon a credible product advantage, but it also creates a demanding standard. A system that worked inside a company’s own operation must now migrate data, reconcile balances, preserve decision histories and support customer teams with different processes. Mortgage borrowers cannot pause their payments while an implementation team finds an edge case.

Customer Scale Turns Proof Into Migration Work

Valon reports more than $200B in mortgages already running on ValonOS and more than $200M in contracted annual recurring revenue signed in the first six months after opening the platform to external servicers. The company also says institutions responsible for 1 in 6 outstanding U.S. mortgages are under contract. Those figures are company-reported and have not been presented as independently audited results.

Carrington and ServiceMac are live. Rithm Capital’s Newrez partnership announcement says Newrez plans to begin moving more than 4M homeowners onto ValonOS in 2027. Each customer adds evidence that the market wants a replacement for legacy servicing infrastructure. Each one also adds a migration where data quality, regulatory controls and customer service have to survive the handoff.

Valon’s $150M Series D, announced in October 2026 at a $2.3B valuation, gives the company more capital for that work. Ribbit Capital joined as a new investor and Andreessen Horowitz participated again. The round funds product development and hiring across engineering, product, deployment and go-to-market, but the real scorecard will be successful customer migrations.

Leadership and Culture Favor Domain Judgment

Valon’s leadership spans technology, operations and regulated finance. Andrew Wang leads the company as CEO and co-founder, with Linda Du as President and COO and Jonathan Hsu as co-founder. The current team also includes Brian McGrath as Executive Chair, Hongxia Zhong as Head of Engineering, Jake Mintz as Chief Product Officer, Marisa Rackson as Head of People and Ariel Brito as Chief of Staff.

The company’s most revealing culture decision is a temporary restriction on AI access for most new non-engineering hires. New employees first learn mortgage servicing closely enough to question and verify model output, then gain access when a manager believes that judgment is ready. Engineers retain AI access from day one. For an AI company, the rule sounds almost impolite. That is why it is useful: Valon is saying that automation cannot replace the experience required to catch a confident error in a regulated workflow.

Other internal practices reinforce that thesis. Valon says it turns recurring code-review lessons into reusable rules, operates Nicolebot as a reviewed company-knowledge agent and built DeploymentOS to manage enterprise implementations. The company is not only selling institutional memory as a product feature. It is trying to make institutional memory part of how the organization works.

Hiring Shows Where the Pressure Is Building

Valon’s current jobs board spans engineering, mortgage product strategy, deployment, product design, operations and sales. An Andreessen Horowitz portfolio profile listed 25 open roles when reviewed, including mortgage product strategists, deployment leaders and engineering managers across New York, San Francisco and remote options.

That mix maps directly to the commercialization challenge. Engineers have to keep the system correct and extensible. Mortgage specialists translate operating rules into software. Deployment teams move customers without losing data or process history. Sales and product leaders have to make a large infrastructure purchase legible to institutions that know switching costs are real.

Valon describes the work as high ownership across software, operations and regulation. The company’s careers page is blunt that people who want cleanly bounded assignments or a ready-made playbook may not enjoy it. That candor is more informative than generic culture language because it matches the product: Valon is hiring people to sit inside the messy handoffs its software is supposed to organize.

What Valon Signals for Regulated AI

Valon represents a broader shift in enterprise AI. In regulated industries, the most valuable agent may depend less on a novel interface than on rebuilding the system that stores facts, permissions, approvals and exceptions. An agent can draft a borrower response quickly. A servicer still needs to know whether the payment moved, which rule applied and who can defend the decision months later.

Valon plans to extend the architecture into commercial, personal, auto and student lending, but that remains an ambition. Mortgage servicing is already a large enough test. The company now has to show that an operating model learned through one servicer can become reliable infrastructure for many.

The attraction is easy to see. Valon built the hard system before it sold the easy story. The next chapter is less romantic and more consequential: enterprise migrations, customer-specific rules and millions of financial records that all need to arrive intact.