Stackpoint co-founds and funds agentic AI companies in complex, high-barrier industries where AI-native systems can fundamentally change how work is done. We are highly selective by design, launching and investing in only four companies each year.
This article outlines why we see such a compelling opportunity in non-agency mortgage lending, how we evaluate markets like this, why we co-founded LoanLight with Max Klein, and why it stood out as a clear fit for our investment thesis.
The Opportunity: A $500B Market Trapped by Old Systems
Mortgage lending is changing, even if the systems that support it have not.
Borrower profiles today look fundamentally different than they did when most underwriting infrastructure was designed. Self-employed professionals, gig workers, real estate investors, and asset-rich borrowers now make up a growing share of demand. Loans increasingly fall outside the narrow constraints of traditional agency programs.
As a result, non-agency lending—loans originated outside the government-backed ecosystem—has grown rapidly. This includes non-QM, jumbo, DSCR, and other alternative products that better reflect how people earn income and build wealth today. Total U.S. mortgage originations are projected to reach roughly $2.2 trillion by 2026, with non-agency volume representing $500B of that market.
Within this broader category, non-QM has moved from the margins to the mainstream. Annual non-QM origination alone now exceeds $100B, supported by strong investor appetite and attractive risk-adjusted returns. Securitization volumes reached record levels in 2025, signaling growing institutional confidence in the asset class.
What has not kept pace with this growth is the infrastructure. Despite its scale, non-agency underwriting still relies on workflows built for a different era: manual document review, fragmented tooling, guideline interpretation living outside systems of record, and judgment applied inconsistently across files.
This is most visible in non-QM, where borrower complexity and documentation variability make the problem impossible to ignore. But the underlying issue is not specific to one product type. Across non-agency lending, the absence of standardized underwriting infrastructure leads to predictable outcomes—longer cycle times, higher origination costs, and greater downstream risk.
Unlike agency lending, which benefits from shared data standards and a widely adopted Automated Underwriting System, non-agency lending lacks a common system of record. Without a consistent, continuously validated view of the loan, the market compensates with human effort instead of systems.
The same loan is often underwritten multiple times—by the originating lender, by buyers or investors, and again during post-close quality control or diligence. Each party revalidates documents, recalculates income, and rechecks eligibility logic because there is no trusted source of truth.
How AI Changes Everything and Why Now
There are many things AI is good at, and many things it isn’t. The most durable AI companies aren’t built by sprinkling AI onto existing software; they are designed around AI’s strengths from day one.
Agentic AI represents a shift in how software is built. These systems don’t just assist—they decide, act, and adapt inside real workflows. That approach wasn’t feasible a few years ago. Today, it is.
At Stackpoint, we look for problem spaces where this shift can unlock an order-of-magnitude improvement. Non-agency underwriting is a clear example. The work is document-heavy, rule-dense, and repetitive—yet highly consequential. It is exactly the kind of environment where persistent, specialized AI agents outperform static tools and manual review.
Non-QM is where this pain is felt most acutely today. The loans are complex, the margin for error is thin, and inefficiencies are costly. But the same dynamics increasingly apply across the broader non-agency market as loan complexity rises and private capital plays a larger role in origination.
How Stackpoint Evaluated the Opportunity
When we evaluate an investment, we look at it through several lenses. In LoanLight’s case, four stood out.
Market structure
Non-agency lending is large, growing, and structurally underserved. While non-QM highlights the problem most clearly, the absence of a standardized underwriting system affects jumbo, investor, and other alternative loan types as well. There is no equivalent of DU or LP outside the agency world, and that absence creates real economic friction across lenders, investors, and originators.
Workflow depth
LoanLight operates where the work actually happens: inside the loan file, inside the LOS, and across the life of the loan.
Architecture
LoanLight is AI-native by design. Its system is composed of specialized agents, each responsible for a core underwriting function—document health, data validation, application consistency, and property analysis. These agents operate continuously and in coordination. They don’t run one-time checks; they maintain a living, validated view of the loan as it evolves.
Wedge and expansion
The company starts with pre-underwriting quality assurance—a universal pain point that is easy to adopt and immediately measurable. While non-QM is the initial proving ground, the platform naturally extends to other non-agency loan types that share the same structural challenges. From there, it expands into eligibility, pricing, diligence, and investor workflows, all built on the same intelligence layer.
What We Are Excited About
We believe LoanLight is a significant value unlock in non-QM, where inefficiency and risk are most visible today. Clean files enable faster underwriting. Faster underwriting enables tighter pricing and higher throughput. Consistent decisions reduce downstream risk. Everyone in the ecosystem benefits.
But the opportunity does not stop with non-QM. The same infrastructure gap exists across the broader non-agency market, which now represents hundreds of billions of dollars in annual origination and continues to grow as borrower profiles evolve and private capital expands its role.
That compounding effect—starting where the pain is greatest, then expanding across adjacent loan types—is what makes systems like this defensible and capable of creating durable, long-term value.
