An Enterprise Standard in Federal Student Debt Repayment, Built on a Self-Checking AI Loop

The future of a company isn’t the model it rents. It’s the loop it owns. Here’s ours — running in the one place you can’t fake precision.

There’s a shift happening in how companies get built, and most of the conversation about it is still abstract. Let me make it concrete, because we’ve been living inside it.

For most of software’s history, holding a hard standard required scale. If you wanted to model something genuinely complex — exactly, and keep it exact as the ground moved under you — you needed an army of engineers and the budget to match. Small teams shipped “good enough.” That wasn’t a failure of will; it was the physics of the work.

That physics changed. The new unit of building isn’t a bigger team, or a better model you rent for a month. It’s a loop: people and AI working together, where human judgment sets the direction and the system gets sharper every time it runs. Point compute at a problem with no one steering, and you get motion without progress — compute running in circles. Put disciplined people at the center, directing it, checking it, feeding what they learn back in, and the expertise compounds. The loop becomes the most valuable thing the company owns.

We decided to test that idea in the least forgiving place we could find.

Why federal student loans

More than 42 million Americans are repaying federal student debt, into a system that was just rewritten — the new RAP plan, the end of SAVE, the collapse of the old income-driven menu — with more changes coming. The people inside this are physicians, dentists, public-service attorneys, dual-income couples carrying six figures while trying to buy a home and raise a family. When the math decides whether a household gets its footing back, “about right” isn’t a standard — it’s a liability. A recommendation built on a rule that changed last quarter, or off by a quiet rounding error, isn’t a small miss. It’s the difference between a plan that helps and one that costs a family thousands.

It’s also brutally hard to model. Multiple repayment plans over decades, each with its own formula. Filing status — jointly versus separately — swinging the answer. Household income, family size, the timing of contributions. Forgiveness tracks like PSLF layered on top. And a federal rulebook that refuses to sit still. Modeling all of that, correctly, and keeping it current as the rules move, used to require an institution — and even institutions settled for approximations, because perfection across that many moving parts, on a moving target, was out of reach.

That’s the bar we wanted to clear with a small, focused team and a loop.

What the loop actually is

Three parts, and the order matters.

First, the expertise is the IP. We encoded the federal repayment rulebook — RAP, the IDR formulas, PSLF, the forgiveness clocks, the filing-status logic — into the engine and into a second, independent reference implementation. That domain knowledge is ours; it doesn’t live inside any one AI model we could lose tomorrow. Swap the model underneath, and the veteran — the accumulated understanding of how this system actually behaves — stays.

Second, humans direct it. This is not automation you point at a problem and walk away from. People who know this domain cold set the goals, make the judgment calls, and refuse to let “the AI probably got it right” stand in for “we confirmed it.” AI doesn’t remove the rigor. Directed well, it lets a lean team reach a rigor that scale used to chase and miss.

Third — the part I’m proudest of — the loop checks its own work. We keep an independent reference model of the federal math whose only job is to disagree with us. It’s our private evaluation: not a generic benchmark, but a check against the outcomes that actually matter — the federal repayment figures an advisor puts in front of a client. Every federal calculation the engine produces gets checked against it, to the cent. When the two agree, we ship. When they don’t, we stop. Every disagreement we resolve makes the check sharper and the knowledge deeper. That’s the compounding part: the loop is a hill-climbing machine, and the climb is ours.

What it produces — and what it doesn’t claim

These are pro forma projections. The future they model — income growth, inflation, the choices a household makes — is set by assumptions the advisor controls, not foreseen by us; no honest tool predicts a person’s next 25 years to the penny. What we hold exact is the math itself: given a set of assumptions, the federal repayment figures — the IDR and RAP payments, the plan comparisons, the forgiveness amounts — are correct to the cent and current with the rules. The advisor owns the assumptions. We own the arithmetic. And software is imperfect — we’ll fix what we find, like everyone who ships.

That distinction is the whole point for the people who carry the weight of these numbers. For the advisor, it’s what shows up in the meeting: RAP versus the new Standard for a high earner, forgiveness tracked, a household re-modeled against the OBBB overhaul — figures they can stand behind. For a firm’s risk and compliance teams, it’s a control they can inspect: an independent model that exists to check our math, and a clear line between the arithmetic we own and the assumptions the advisor sets.

Why this is the part worth watching

The precision is the proof. The loop is the point. An advantage built from a compounding loop — domain IP, human direction, a private eval that sharpens every cycle — is genuinely hard to replicate, and it doesn’t depend on whichever model is best this quarter. The companies that build this early, in domains that matter, won’t be overtaken by the next model release. They’ll be the ones the next model plugs into.

We’re proud to put this in the hands of advisors — and proud to be a dependable part of the stack a serious firm builds on, the kind of component a diligence team can examine and an advisor can rely on without a second thought. We didn’t build a feature. We built the loop, and pointed it at a problem where being right changes a family’s life.

The old way was scale, or settle. The new way is a loop that compounds. That’s the bar we build to — and the reason we built the company at all.