AI agents are moving beyond customer-service chatbots and productivity assistants and into one of the most complex areas of financial services: mortgage lending.

Vesta, an AI-native software company that provides technology for mortgage loan origination, has raised $30 million in a new funding round led by Conversion Capital.

The round includes participation from several of Vesta's customers, including Pennymac and New American Funding, as well as Citi Ventures and Andreessen Horowitz.

The funding gives Vesta more capital to expand its AI-agent platform, develop new products and compete for a larger share of the U.S. mortgage technology market.

The company says demand for its technology has accelerated significantly, with revenue reportedly growing 12x year over year and its platform now supporting lenders originating more than $100 billion in loans annually.

Vesta wants AI agents to handle the mortgage workload

Mortgage origination is a highly document-heavy and process-intensive business.

A single loan can require borrowers, loan officers, processors, underwriters and other professionals to work through large amounts of financial and personal information.

Vesta's approach is to break that process into individual tasks that can be handled by software and increasingly by AI agents.

Its current platform allows lenders to automate work from the application stage through funding, with agents able to interpret documents, use mortgage-specific tools and coordinate tasks across teams and external providers. Vesta says its system maintains an auditable record of loans, borrowers, properties and documents.

The goal is not necessarily to remove humans from the mortgage process.

Instead, the company wants lenders to deploy multiple AI agents alongside human employees, allowing people to supervise the work and gradually give agents responsibility for more tasks.

From human approval to greater autonomy

Vesta CEO and co-founder Mike Yu says lenders can determine how much autonomy an AI agent receives.

A lender might initially have an employee review everything an agent produces.

Once the organisation becomes comfortable with the system, the agent can handle a larger share of the workload independently.

That creates a gradual path toward automation rather than requiring a lender to hand over an entire workflow to AI on day one.

It also reflects a broader shift in enterprise AI.

Companies are increasingly moving from AI that generates content to AI that executes processes.

In mortgage lending, that could mean an agent checking documents, identifying missing information, evaluating conditions, updating a workflow or preparing a file for human review.

Vesta says some lenders are even using its agents in underwriting workflows, although the lending companies remain responsible for the decisions.

Why mortgage lending is an attractive target

The economics help explain why companies are pursuing this technology.

According to the figures cited by Vesta, closing a mortgage in the United States takes roughly 40 days and costs around $11,000 per loan.

A significant portion of that cost comes from human labour and the time employees spend waiting for another person to review or complete a task.

If AI agents can reduce those delays, lenders could potentially process more loans without increasing headcount at the same rate.

That could become particularly valuable when mortgage volumes fluctuate.

Instead of hiring large numbers of employees during busy periods and cutting staff when demand falls, lenders could use software agents as a more flexible layer of operational capacity.

Vesta itself describes this as turning fixed headcount into more elastic capacity that can expand or contract with demand.

New American Funding is already preparing for the shift

The company is not simply pitching the technology as a future concept.

New American Funding announced in July that it had selected Vesta as part of an effort to modernise its origination, processing and closing workflows.

The phased rollout is scheduled for 2027 and is intended to improve efficiency while making the mortgage experience faster and more transparent for borrowers.

Verus Mortgage Capital also went live on Vesta in July 2026, replacing its legacy loan-origination system.

Vesta says its task-based architecture allows individual pieces of a loan to be routed either to a human or an AI agent, including workflows involving specialised credit decisions and exceptions.

Those deployments are important because mortgage technology is not a simple environment.

Lenders have to deal with regulations, documentation, underwriting rules, fraud controls and highly individual borrower circumstances.

The AI has to be auditable

Putting AI into mortgage underwriting creates an obvious concern:

What happens when the system makes the wrong decision?

A conventional chatbot can give someone a bad recommendation.

An AI system involved in lending could influence whether someone gets approved for a mortgage, how quickly the loan closes or what conditions the borrower has to satisfy.

That makes traceability critical.

Vesta says actions and reasoning associated with AI decisions are recorded so lenders can audit what happened and maintain compliance.

Human responsibility also remains important.

The lender, rather than the software provider, remains responsible for the underwriting decision.

That distinction could become increasingly important as regulators and financial institutions work out how existing rules apply to increasingly autonomous AI systems.

Better AI models are making more ambitious agents possible

Vesta's founders say the company's strategy has changed as AI models have improved.

Earlier generations of AI were not reliable enough to handle the long, complicated sequences of tasks required in mortgage lending.

The company initially focused heavily on building the underlying data architecture needed to automate mortgage processes.

Now, Vesta believes newer models are capable of following detailed instructions over longer periods and completing more complex workflows.

The company specifically pointed to Anthropic's Claude Sonnet 4.5 as a major improvement for its use case.

That highlights an important trend across enterprise AI.

The advancement of foundation models is not simply producing better chatbots.

It is enabling startups to build specialised agent layers on top of complex industries.

Vesta is entering a crowded AI mortgage market

Vesta is not alone.

Traditional mortgage technology providers such as ICE Mortgage Technology remain deeply embedded in the industry, while newer AI-native companies are targeting specific parts of the lending process.

Other startups are focusing on areas such as AI-powered mortgage sales, borrower communications and underwriting.

The competition could ultimately force mortgage software to become more modular and automated.

Vesta's advantage is that it is trying to operate across the underlying loan-origination workflow rather than focusing on a single interaction.

Its current platform is designed to serve as a system of record while allowing both deterministic rules and autonomous agents to operate within the same environment.

The bigger fintech story

Vesta's funding round points to a broader change in financial technology.

The first wave of fintech focused heavily on putting banking and financial products online.

The next wave is increasingly about automating the work behind those products.

AI agents could eventually handle parts of lending, insurance, payments, compliance, fraud detection and financial operations.

But financial services are also where the risks become much higher.

An AI agent processing a spreadsheet is one thing.

An AI agent involved in deciding whether someone can borrow hundreds of thousands of dollars is another.

That means the winners in financial AI will likely need more than powerful models.

They will need strong audit trails, clear human accountability, secure data infrastructure and predictable behaviour.

Vesta's $30 million raise shows where the AI-agent race is heading: into the infrastructure behind financial decisions.

The interesting part is not that AI can read mortgage documents.

It's that lenders are increasingly willing to let AI participate in the actual workflow that determines how those documents are processed.

That could reduce costs and accelerate mortgage approvals, but it also raises a fundamental question about financial AI:

How much decision-making should we allow machines to handle when the outcome can determine whether someone gets a home loan?

The technology may be ready for greater autonomy.

The financial industry still has to decide where the human line should remain.