About Adeft
Adeft Inc. is building an AI-native processing department for wholesale mortgage brokers. Our agents will do the file work. A licensed processor will approve every outbound action.
We have been the originator working the file after hours. We have been the self-employed borrower whose income got a second look. Before Adeft, we spent years building systems that do this kind of work. The file is where all three meet. That is where we build.
A founder replies within one business day. No sales team in between.
Founders
One has originated loans. One has automated closings. Both have shipped into regulated work.
Co-Founder and CEO
Three-time founder. Began as a mortgage loan originator and helped open and run a branch office for a multistate brokerage. Then 15+ years building data and workflow products, most recently agentic AI for the enterprise.
Co-Founder and CTO
Built AI-first title insurance and closing automation at Doma. Led engineering for Ford's greenfield compute platform.
Advisors
Sam Sharp
EVP, CrossCountry Mortgage
Formerly EVP at Guaranteed Rate. $2B+ in personal funded volume.
Gary Sumner
Two-time security and cloud founder
Formerly at Sophos, ServiceNow, and Ford.
Joe Kleinschmidt
Founder, multiple exits
Built and sold AI companies in B2B and B2C.
Advisors serve in a personal capacity, are not compensated by loan volume or referrals, and do not see borrower data or file records. Affiliations are listed for identification only and do not imply endorsement by any employer.
Why processing first
In wholesale, the broker owns the relationship and the judgment. The processing desk owns the hours. We started there on purpose.
Self-employed and complex-income files punish borrowers hardest, and brokers know them as the files their shop is best at. That is where we intend to keep getting better.
Document chasing, classification, income analysis, conditions, letters of explanation, status updates. Each one is time a desk spends not deciding. That time can be measured, and we build against it.
Processing has long been supervised work, done under a licensed company's name where the state requires it. Our agents will prepare the work; a licensed processor will approve it before anything goes out. We did not have to invent the review step. The industry already had one.
Delegation should not mean losing sight of the file. Every file carries a record of who did what and when, and your loan officers can see it at any time.
How we work
Zero outbound actions without a licensed processor's approval. No autonomy mode, and no exceptions for routine status updates. Every action is timestamped, attributed, and exportable.
Nothing ships on our opinion. Once we are open, a licensed processor will try every change on real work, under the processing company's rules, before it stays. If it does not save time on the desk, it goes back to the bench.
One test we hold ourselves to: does a change save a processor time on the file without touching the licensed sign-off. When the evidence disagrees with our opinion, the evidence wins.
Long term
Tens of millions of Americans earn self-employed or otherwise complex income. Their files are the ones we chose first, and they set the bar for everything we build.
Over time, we want those borrowers to understand how lenders read their income before they ever apply. Today that means a library of guides we are writing, a newsletter called Building Adeft, and a lot of listening. That is education, not qualification. Adeft Inc. is not a mortgage lender, mortgage broker, or loan processor, and it does not originate, arrange, or offer loans or take loan applications. Nothing we publish is advice about any one borrower's situation. For that, a borrower should talk to a licensed professional.
We are interviewing self-employed borrowers about their mortgage stories. 20 minutes, on video or phone, with a founder. It is research, not an application or advice. We ask before we record. Your name is not given to any lender or broker.
Brokers and processing shops: join the waitlist instead.
Two founding roles, full-stack and machine learning. San Francisco, hybrid. Small team, regulated work, and every change is judged by whether it saves a processor time on a file. Send a founder a short note and something you have built. No cover letter.