Manish Karney

Explorations / Personal AI · Housing

Buying a house with AI as my co-pilot

A first-time buyer from a house-poor background, no inherited playbook, against a process built for people who already have advisors. I ran it with AI — modeling, legal, the offer, and now the renovation.

What it taught me

Now: renovating

AI turns an opaque, advisor-gated process into one a careful first-timer can actually run.

Still open

Where does AI level the field for people without generational knowledge — and where does it quietly mislead them?

Personal AIAccessJudgment

The stakes

A process built for people who already have advisors

I bought my first house from a house-poor background — no inherited playbook, no family lawyer to call, no parent who had done this before. The process assumes you have those people. Each step hands you a form to sign and a professional to trust, and quietly expects you to already know which questions to ask.

That is the part that struck me: the people with the least slack get the least guidance, exactly where the stakes are highest. So I did the thing I do at work. I refused to treat the purchase as a black box, and took it apart instead — with AI as the advisor I did not inherit.

AI was the first advisor I could afford to inherit.

What I took apart

Six opaque sub-processes, each made legible

A home purchase is not one decision; it is six tangled ones, each with its own jargon and its own way to lose money quietly. I worked through them the way I read a system I do not yet trust — asking the model to explain, to run the numbers, and to show its work, then checking it against people who do this for a living.

Modeling
Affordability, total cost of ownership, and the real monthly number behind the listing price.
Legal
Disclosures, contingencies, and contract clauses, read and explained before I signed anything.
The offer
Comparables, strategy, and the number to bring — reasoned from evidence, not guessed under pressure.
Renovation
Scope, bids, and sequencing for the work the house needs now versus the work that can wait.
Mortgage
Rate, points, and refinance math, re-run every time the numbers moved.
Financials
Cash flow, reserves, and the ongoing tracking that keeps a stretch purchase honest.

The journey

Where the model helped, and where the call stayed mine

AI rode along on every step — translating the language, running the numbers, surfacing the question I did not know to ask. But the shape of the thing matters: one step is irreversible, and no model should make it for you.

The model informs the offer → a person still makes it

Fig. 01The purchase, end to end — AI on every step, the offer owned by a human

AI-assistedmy call

The offer is that step. The model can price it, stress-test it, and talk me out of a bad one — but signing it is a human, accountable act, made with incomplete knowledge about a place I want to live. AI got me to the decision well-prepared. It did not get to make it.

What it taught me

Where AI leveled the ground, and where it stopped

Where it helped most was translation: turning advisor-gated language into something a careful first-timer could reason about. Modeling, comparables, the meaning of a clause — the work that used to require people I did not have.

Where it stayed human was judgment: trusting a contractor, reading a neighborhood, accepting an irreversible risk, and knowing what the model did not know. The skill was not offloading the decision. It was using the model to reach the decision better informed, then owning it.

The same conviction runs through the rest of my work: put the capability where the person with the least slack can reach it.

This is the healthcare thesis rehearsed on my own life: frontier tech made tractable for the person a system serves last, with a human in the loop where it counts.