Meta released Muse in September, a personal AI agent that can research, use websites and complete tasks on your behalf. Apptopia estimates it generated 5.8 million downloads across the U.S. and Canada in its first 20 days. Nearly half its early U.S. users were 46 or older.

I used it last week to help me find a used car. I had specific criteria, so I asked Muse to compare reliability, maintenance costs and insurance premiums, then recommend a few models. It identified models that fit, searched local listings and compared prices, mileage and insurance rates. Within 10 minutes, it brought back five options for me to consider. That could have been hours of clicking through listings.

Then I asked it to negotiate my internet bill. It started an online chat with my service provider. When the provider pushed back, I told Muse to try harder. It went back and persuaded the representative to knock $10 a month off. Less than 30 minutes later, the discount confirmation landed in my inbox.

Muse still needs supervision, and connecting it to your tools means being careful about what you let it access. But the appeal is easy to understand: describe what you want and delegate the work of finding it.

Mortgage shopping has plenty of that work.

The borrower can choose before you meet

For decades, Google and Meta have monetized access to consumer attention. Lenders pay to appear when someone searches or signals interest. The consumer still has to investigate the offer, compare alternatives and make contact.

An agent can do more of that evaluation before the consumer ever reaches a lender’s website. By the time a loan officer enters the conversation, the shortlist may already exist.

A lender’s own chatbot starts with someone who has already found the lender’s website. A consumer’s agent shows up much earlier and can decide which lenders to approach. Improving your chatbot won’t necessarily earn you a place in that earlier conversation.

Muse isn’t the only product moving in this direction. Grok Bot, positioned more toward work and teams, also uses tools and websites to complete assignments. Amazon’s Buy for Me experiment lets an agent purchase from another brand’s website after the customer confirms.

Amazon, Uber and Zillow have already demonstrated the appeal of reducing consumer effort. Agents extend that convenience into the research and coordination consumers still do themselves.

Trust will develop unevenly. A Gartner survey of 322 U.S. consumers, conducted before Muse launched, found greater willingness to let AI narrow retail choices than make purchases. It wasn’t mortgage research, but the distinction matters.

A borrower can keep the final decision while delegating enough shopping to change which lenders get considered. Lenders don’t need to wait for autonomous closings to take that shift seriously.

Is this mortgage’s “Zillow moment”?

By 2016, Zillow Group’s Premier Agent revenue had reached $604.3 million. Zillow had built a substantial business around where consumers began their search and how professionals reached them. For mortgage executives, the lesson is about access. Control over the customer relationship can shift even while the transaction still requires established professionals.

Personal agents could gain influence over an earlier decision: which providers deserve the consumer’s attention at all. A lender might lose an opportunity without knowing the borrower was even looking.

Zillow is already adapting. It introduced its ChatGPT app in October 2025, bringing listings into the conversation while directing important next steps back to Zillow. A new interface can give an established intermediary another way to reach consumers.

Who owns the agent matters. So does who pays for it.

Can the biggest lenders buy their way onto the shortlist?

Google has announced tests of sponsored formats within AI Mode recommendations. Advertising can follow consumers into an experience where software helps them decide.

That doesn’t establish that Muse sells mortgage recommendations. Meta says Muse doesn’t share conversations or data in users’ virtual computers with its advertising systems.

Advertising accounted for roughly 98% of Meta’s revenue in 2025. That doesn’t tell us how Muse will make money, but it gives lenders a reason to watch closely. I would be surprised if access to consumers actively shopping for a mortgage didn’t attract commercial interest. The question is how that interest will shape what the agent recommends.

The consumer, meanwhile, expects the agent to work on their behalf. If commercial relationships narrow the options it considers, the consumer may never see what was excluded. Knowing how a recommendation was selected matters when you’re committing to years of payments.

Larger lenders could gain an advantage even without paying for placement. They can fund integrations, keep pricing current and make their systems easier for agents to work with. That investment could influence which providers an agent can successfully evaluate.

Smaller IMBs could benefit too, if agents can identify and verify a particularly good fit for a borrower. That depends on reliable information, competitive offers and how the platform selects recommendations. A new channel doesn’t automatically level the market.

Before committing money to an agent channel, executives should understand who can appear, what influences selection and what the lender is paying for. A stream of inquiries may be valuable. But relying on another platform to reach suitable borrowers can turn a useful source of business into an expensive dependency.

A low rate is only the beginning

Imagine a buyer asking an agent to find homes and financing within a monthly budget, with limited cash for closing and a target move date.

The agent could research properties and approach lenders with an organized request. The lender then needs to establish whether the borrower qualifies and provide a comparison worth trusting.

A headline rate with substantial points may leave this buyer short of cash. Taxes and insurance may put an otherwise affordable house beyond the monthly budget. A neat table can conceal very different assumptions.

An experienced loan officer still has meaningful work to do: resolve uncertain income eligibility, explain tradeoffs and assess whether a closing timeline is realistic. The opportunity is to start that conversation with the borrower’s research and priorities already understood.

If the borrower has to start over, much of the convenience disappears at your front door.

Build a lender my agent can work with

Lenders can start preparing before the platforms settle on how to make money from this channel.

Start with what an agent can reliably learn about your business: where you lend, who qualifies and what your specialties actually mean. If that information is hard to find or inconsistent, you make it harder for an agent to explain why your offer belongs on the shortlist. Rates need assumptions, relevant costs and dates, with a clear route to a current quote. Clear websites, consistent documents and responsive people are useful starting points. You don’t need a major integration project to begin.

Preserve the borrower’s progress. Provide a secure way to share organized information and questions. Ask for what’s missing. Let the loan officer continue the conversation instead of reconstructing it.

Keep permissions specific. Researching products, submitting an application and authorizing a credit inquiry are different actions. “Find me a mortgage” shouldn’t be treated as permission for all three.

Give an agent a hypothetical borrower your business should serve well. Ask it to compare you with competing lenders. Does it find your advantage? Can it explain it accurately? Where does the process break? Keep the exercise within research and comparison, without submitting applications or triggering credit activity.

Watch where it misunderstands a product, loses pricing assumptions or reaches a dead end. Give someone responsibility for fixing the journey across marketing, operations and technology. If the agent explains a competitor’s offer more accurately than yours, investigate why. You may find a problem your human customers already encounter.

Measure the time to a usable answer and how often the borrower must repeat information. As actual borrowers start using the channel, track acquisition cost per funded loan and margin. More comparison could pressure pricing while better handoffs reduce handling costs. Inquiry volume alone won’t tell you whether you’re winning.

Mortgages are consequential, infrequent purchases. If agents create more corrective work than they save, adoption could stall. If they improve, borrowers have a strong incentive to use them. Clearer offers and better handoffs help customers under either outcome.

When Muse brought back my car options, I still had a decision to make. Hours of research had already been done for me.

Your next borrower may arrive expecting the same. How much of that work will your process make them do again?

Reuven Gorsht helps leaders in lending and financial services get fluent enough with AI to lead the change instead of approving it. He is the co-founder of The Variable, an AI advisory firm, and the CEO of Deeded, a virtual real estate closing platform. MoveSnap, which he co-founded, was acquired by RBC in 2018. Before that he spent more than a decade at SAP, where he led customer strategy. He is a keynote speaker and writes The Human Variable on Substack.