How an AI-Native Private Capital Operator Spends a Week

What does it actually look like to run a real-estate private equity fund alongside a portfolio of operating businesses staffed by AI agents?

I get the question more than any other from operator-builders trying to picture the model in practice. The answer most people imagine is some sort of frenetic many-hats juggling act. The actual shape of my week is closer to the opposite — long, protected blocks of focused work — and I think the structure is itself the most interesting thing about being AI-native. So this is a walkthrough of where my hours go.

A few caveats up front:

This is roughly the seventy-fifth percentile week. Bad weeks (an LP question that takes twenty hours to answer well; a property emergency; a CAD migration that breaks an enrichment) consume the calendar.
I’m writing this in January 2026, about a year into running Uplift Capital and a few quarters into running the broader operating-portfolio model. The cadence is still settling.
The numbers and time blocks are real, not idealized. I rounded.

The shape of a week

Per Beeminder + a calendar I actually keep, my work week averages out to about sixty hours, broken roughly:

Block	Hours/week	Where it goes
Uplift Capital — primary mandate	28	Partner work, deal flow, asset management
Uplift OS / infrastructure building	8	The system that runs Uplift
AI-managed operations review	5	Approving / adjusting agent-run businesses
Writing + reading + research	8	Essays, the letter, books, papers
Outreach + relationships	5	Cold-but-warm emails, calls, partner conversations
Community + farm + family	(separate)	Weekends + early mornings
Protected (no work, no screens)	6	Saturday afternoon onward

The single most important number on that table is the 28 hours on Uplift Capital. The fund is the principal mandate; the partners’ capital and my fiduciary attention live there; and that ratio — close to half my work week — is the answer to anyone wondering whether the broader model dilutes attention.

Now the actual blocks.

Monday + Tuesday — Uplift Capital

Two full days of partner-letter drafting, deal flow, asset management, and capital markets.

The work that’s mine on these days:

Investor relations. Reading partner emails, drafting responses, prepping for monthly partner calls, walking through quarterly numbers. This is the work that has to be done in my voice, with my judgment. AI helps draft and organize, but the words on every letter that goes out are mine.
Deal flow review. Whatever teasers came in over the weekend get triaged Monday morning. The live deals from the prior week get an end-of-week status. Acquisitions team check-ins. Broker conversations.
Asset management. Property-manager status. Anything flagged in last week’s Uplift OS run that needs a human read. Capex pipeline. Refi/leverage decisions.

The work AI does for me on these days:

Drafts every LP letter section from the prior month’s GL data and operating metrics, then I rewrite. Compositional editing, not blank-page writing.
Scores every incoming teaser against our deal box and surfaces the ones worth a closer read. Bottom-quartile teasers get a polite decline draft I just send.
Anomaly-flags the portfolio. When a fee charge looks unusual, when a capex line item is trending past budget, when a vacancy spike doesn’t match the market, the system tells me before the property manager does.
Pre-reads every relevant document. Before a broker call, an agent has read every document in the data room and written me a one-pager.

That last one is the thing that compounds the most. Twenty years ago, a deal team had three associates do that work. Today I do the work of three associates in about ninety minutes per deal, and the quality bar is closer to a senior than a first-year. The economics of running a small fund with institutional discipline don’t work without it.

Wednesday — Builder day

Wednesday is for the system that runs the operation. Uplift OS gets its weekly maintenance pass. New ingestion sources I’ve been queueing get wired in. Any agent-output drift from the prior week gets debugged.

A typical Wednesday looks like:

Morning, ~3 hours. Review the prior week’s Uplift OS classification work — every GL row tagged “rule” passes silently; every row tagged “llm” with confidence below 0.7 lands in a queue I review and either approve or push back to the prompt as a counter-example. Then I push the cleaned-up prompt to production.
Midday, ~2 hours. Whatever the next ingestion priority is. This week it’s a property manager who started exporting in a slightly different format; I update the per-source parser.
Afternoon, ~3 hours. The longer-cycle infrastructure work — schema changes, query optimizations, dashboard tweaks for the partner-call deck.

I’m the only person who builds Uplift OS, and I think that has to remain true. The system embeds my opinions about what “anomalous” means in our portfolio. The day I outsource that, the system stops being the operator’s instrument and becomes a vendor product. Vendor products don’t catch fee leakage that lives in the management agreements. My instrument does.

Thursday — AI-managed operations review

This is the day the broader operating-portfolio model justifies itself.

I own a small portfolio of operating businesses in adjacent verticals — the kind of companies a family office might hold inside a holding-company wrapper. Each runs on delegated AI agents, not on my time. They have monthly autopilot cadences: content generation, customer interaction, ops cadence, reporting, billing where applicable. The agents do the work. My job is to review what they did and approve what they queue.

A Thursday looks like:

Read the autopilot reports. Each business produces a structured digest of what happened in the prior week — content shipped, customers handled, money in, money out, anomalies flagged. I read all of them. They take maybe forty minutes total.
Approve queued exceptions. Things the agents flagged for human attention: a content piece that touched a regulatory question, a customer contract that exceeded the approval threshold, an ops decision the agent wasn’t confident on. Each is presented with the agent’s reasoning. I either approve, redirect, or escalate. Usually a couple hours of total review work.
Adjust agent prompts. When the agents drift — when content quality dips, when the customer-handling tone goes wrong, when the ops cadence misses something obvious — I update the prompts. This is the operating-system work for the OpCos, equivalent to what Uplift OS is for the fund.

Total Thursday time on the OpCos: about five hours. They throw off cash; that cash goes back into Uplift Capital, the farm, or other long-term assets. The agents handle the work that, in a non-AI-native version of the same business, would require a small team.

This is the part I think people most underestimate. The economic surplus of running businesses on AI agents instead of human teams accrues to the principal — the owner — not the labor. A few small businesses run this way produce real cash flow with very little of my time, and the cash flow funds the work that requires my attention.

Friday — Writing, reading, and outreach

Friday is the day I push back against my own operator-builder tendency to never write or distribute.

Morning: writing block. Two to three hours on whatever I’m publishing next. This essay was written across three Friday mornings.
Midday: reading. A couple of hours with a book, a paper, or whatever’s caught my attention in the prior week. I keep a running list. I’m currently reading Howard Marks’s quarterly memos, a TWDB whitepaper on the Trinity aquifer, and Christopher Alexander’s A Pattern Language (yes, the architecture book — yes, it applies).
Afternoon: outreach. Five emails to family-office, RIA, attorney, or operator contacts I’ve been meaning to circle back with. Sometimes a Calendly slot or two. The discipline isn’t to close anything; it’s to keep relationships warm.

Distribution is the part of the model I find hardest. My natural inclination is to build, ship, and let things grow organically. The world doesn’t reward that, especially in private capital. So Friday is a forcing function. I won’t always honor it; when I don’t, things slip.

Weekends — community + farm + protected space

I live on a working farm in Gainesville, Texas. Weekend mornings are cattle, chickens, donkeys, fence repair, and whatever else is on the farm list. Saturday afternoons through Sunday I try to keep entirely off-screen — family, neighbors, the Gainesville Economic Development Corporation board if it’s that month, church.

The farm is intentional. It’s the part of the week that won’t run on agents. The model only works if there’s something the agents can’t do, because that’s where the human stays human.

Where this breaks down

A few honest acknowledgments:

Crisis weeks blow up the structure. A property emergency, an LP question that needs twenty hours, a personnel issue at a portfolio company, a system outage. Everything I described above goes out the window. The model has to be flexible enough to absorb those weeks without collapsing.
Distribution still slips. Friday is the most-broken commitment in my week. When something explodes Monday-Thursday, the writing block is what gets cut.
The OpCos require trust in the agents. When agents drift in a way I don’t catch on Thursday review, the businesses do drift. The discipline is constant prompt-tuning, automated quality gates, and the willingness to throttle a business if its agent infrastructure isn’t holding up. Two of my OpCos are throttled right now while I rebuild their content pipelines.
The operator-builder tension is never fully gone. I want to spend more time building. I have to spend most of my time operating. AI dissolves this tension partially — by handling the labor — but not completely. Someone has to think strategically. Today, that person is me.

The lesson

The framing I keep coming back to is that being AI-native isn’t a hat I put on for some tasks and take off for others. It’s a fundamental claim about who’s doing the work.

In a pre-AI version of my week:

Twenty-eight hours on Uplift Capital would be twice as many people doing half as much each, and I’d spend half my time managing.
Eight hours building Uplift OS would be impossible — that work would be quoted at $300K+ from a vendor and not built at all.
Five hours on the OpCos would not be five hours; it would be twenty hours of management overhead, or those businesses wouldn’t exist.

The compression of all that into sixty real hours is the entire premise of the operator-builder. AI handles the labor. I handle the judgment. The cash flow accrues to the principal, not the team.

The week itself is the proof. Sixty hours, weighted heavily toward the principal mandate, with enough left over to write essays, read books, and stay on a working farm. That’s the model — it runs in practice as well as on paper, and it’s how I expect a meaningful share of private capital to operate within five years.