Vesta Raises $30M—AI Now Handles 40% of Its Mortgage Tasks

On October 8, 2026, San Francisco mortgage software company Vesta said in its Series B announcement that it raised $30 million led by Conversion Capital, bringing total funding to $85 million, and that AI agents and automated workflows now complete about 40% of tasks on its platform. That percentage measures completed tasks, rather than mortgages originated or loans approved without human involvement.

In TechCrunch’s interview with co-founder and CEO Mike Yu, he said demand had “exploded in the last year,” revenue had risen twelvefold year over year, and customers decide which tasks to assign agents. Some lenders begin with a person approving an agent’s work before allowing it to handle a share of loans more independently. Lenders remain responsible for underwriting decisions even when they authorize an agent to make one.

Why mortgage businesses joined Vesta’s round

Pennymac and New American Funding were among the investors, alongside Citi Ventures, nbkc bank, First American, FirstKey Mortgage and venture backers. Their participation places mortgage and settlement businesses on both sides of Vesta’s expansion: the company is raising capital from organizations that operate in the market for its software. An investment establishes an interest in that expansion; it does not establish that every investor has deployed agents or delegated the same work.

Vesta sells a loan origination system, the software through which lenders manage applications, conditions and the steps toward closing. An agent embedded in that system can work with the loan record and its existing workflow, where staff also track changes and exceptions. This is the commercial case for agents in origination: much of the work involves reading varied information, applying lender rules and handing a file between parties. The system’s ability to record what happened matters most when an automated action affects a consequential decision.

What the automated task share covers

The production assignments identified with the funding round span application scrubbing, underwriting decisions and closing document review. Application scrubbing happens early, when information entering a loan file needs attention. Underwriting addresses whether a loan meets the lender’s requirements, while closing review concerns material near the end of origination. Treating all of these as tasks makes sense for a workflow count, but their consequences and oversight needs differ sharply.

The reported share combines work done by AI agents with automated workflows. It does not separate an agent’s decisions from rules-based processing, weight tasks by difficulty or disclose how many loans received an agent-made underwriting decision. A routine check and a credit decision may each add one completed task to the total, although they require different levels of review. The figure demonstrates substantial automation on Vesta’s platform; it cannot, on its own, show the experience of an individual lender or borrower.

That distinction helps explain why mortgage origination is an agent market rather than a single automation problem. A loan moves through documents, conditions, outside parties and decisions, creating many opportunities to reduce waiting or manual handling. Automating a handoff may speed the file without changing decision authority. Delegating underwriting goes further: it makes the lender’s controls over the agent and its decision record central to the process.

How lender approval changes during deployment

Customers choose which work to delegate, so agent authority can expand in stages. A lender may initially require a person to approve the output, then permit the agent to handle a portion of loans on its own after observing its performance under that lender’s procedures. That progression is a description of how some customers deploy the software, not a uniform approval policy across Vesta’s customer base.

The operational question changes with the assignment. For an application check, a person may need to resolve missing or inconsistent information before the file advances. For underwriting, the lender needs to know which decision the agent made, what information it used and when a person intervened. Yu described a record of agent actions and reasoning for audit purposes; the usefulness of that record depends on whether it lets the lender trace a particular outcome, especially one challenged by an applicant.

Credit decisions remain the lender’s responsibility

The Consumer Financial Protection Bureau’s guidance says creditors using complex algorithms in credit decisions must still provide applicants with the specific principal reasons for an adverse action. A creditor cannot rely on the complexity of its technology as a reason for failing to explain a denial. If an agent contributes to that decision, the lender needs to connect the factors actually considered with the explanation it gives the applicant.

This obligation puts a practical boundary around autonomy. The lender can decide when an agent works independently and when a person approves its output, but the creditor must still be able to account for the resulting credit decision. Logging that an agent acted is useful for tracing the workflow; explaining an adverse outcome requires a record detailed enough to identify the reasons behind it.

Vesta’s new funding supports wider deployment of its agents within mortgage origination. As lenders extend their authority from routine checks to decisions that affect borrowers, the consequential measure will be whether each lender can show who authorized the work, what the agent did and how an adverse decision was explained.