Why We Build Tools That Don't Replace Judgment
A short note on the design philosophy behind LargeKite — AI augments analyst capacity rather than replacing committee judgment, and why that distinction matters more than the productivity gain.
Most "AI for real estate investment" pitches we see make the same mistake: they imagine AI's job is to score a deal Approve / Conditional / Pass. The verdict.
That's the least useful place to apply AI. The verdict is the part the committee owns. It's also the part where the model is most exposed and most likely to embarrass itself with a hallucinated recommendation.
The valuable application is everywhere else — the work that happens before the verdict. Pre-meeting OM extraction. Standardized risk surfacing across a fixed taxonomy that gets applied to every deal the same way. Drafting the structured sections of the memo so the analyst can spend their time on the judgment sections. Building a queryable institutional memory that lets the committee draw on the last 50 deals.
The pattern is consistent: AI does the parts that are mechanical, consistent, and benefit from being done the same way every time. Humans do the parts that require judgment.
If you do this well, the analyst's deal capacity doubles. The committee gets a more uniform, better-prepared packet every meeting. And the institution builds a memory of its own decisions that compounds over years.
If you do this badly — by asking the AI to render the verdict — you get a system that's wrong with confidence and that destroys the analytical pipeline as junior analysts atrophy.
Both outcomes are entirely a function of where you draw the line between AI work and human work.
The longer version is in AI Workflows for Real Estate Investment Committees.
Share kit
Post this to LinkedIn
Tip: copy the post text first, then click "Open LinkedIn share dialog" — paste into the dialog that opens. LinkedIn auto-renders the URL preview.
More Field Notes
Phoenix Multifamily — The Supply Story Is Decelerating
Phoenix multifamily deliveries peaked at 24K units in 2024. 2026 projects ~13K. The supply pressure that broke rents in 2024-2025 is materially abating — but the price discovery hasn't caught up yet.
The 5.85%-Cap-That's-Really-4.95% Pattern
A short methodology note on the systematic gap between broker pro forma cap rates and recast cap rates on multifamily deals — and what to do about it.
OM Extraction: 47 Seconds, Four Red Flags
Walking through a real OM extraction on a 48-unit Tampa multifamily — the structured output, the red flags surfaced automatically, and where the extraction agent saved an analyst about three hours of work.