Field notes / Buy vs build · Series

Rent vs. Own: The AI Employee Edition

Twice in one weekend I watched operators in my communities get pitched an AI employee for $500 a month, and one of them bought on the spot. Maybe the right call. But nobody at either pitch had a spreadsheet open, so here's mine: what renting costs by month 25, what owning costs, and the one question that separates them.

This weekend I sat through the same pitch twice, in two different real estate communities I'm part of. An AI employee for your business. It answers your leads, works your follow-up, never sleeps, five hundred a month. Slick demo, real capability, and in one of the two rooms, somebody signed up before the call ended.

I want to be careful here, because nothing in this post is a criticism of any vendor. The demos I saw were genuinely good, the category is real, and $500 a month for work that used to take a hire is not an outrageous price. My problem was with the room, not the stage: a group of real estate investors — people who can recite the rent-vs-own math for a duplex from memory — watched a rent-vs-own decision go by and nobody ran the numbers.

So let's run them. Same discipline as every deal you underwrite: cash out over time, what you hold at the end, and what the exit looks like. I sell one side of this trade — installed systems you own — so read everything below the way you'd read a lender's worksheet, and check the arithmetic yourself. All of it is checkable on purpose.

First, notice what's actually being sold

"AI employee" is a great phrase and a misleading category. I've written before about why the employee half doesn't hold up — what I run is a chief of staff, not an employee, because the moment software gets an employee's authority you inherit an employee's audit problem. This post is about the other half of the phrase, the half the pricing page reveals: it's not an employee, it's a lease.

Here's the test, and I'd apply it to any AI service before signing, including anything with my name on it. Call it the cancellation test, the ugly cousin of the vacation test:

For an employee who quits, the answer is: the playbooks stay, the CRM history stays, the phone number stays. For most AI-employee subscriptions, the honest answer is closer to: you keep the memories. Not always — some vendors export everything, and the ones that do will tell you proudly. Which is exactly why it's a fair question to ask out loud, in the Q&A, before you buy.

The arithmetic

Let's price the specific thing that was being sold this weekend: an AI that handles inbound lead response and follow-up for a small acquisitions operation. Two honest ways to get it.

Path A: rent it. $500/month, live this week, vendor maintains everything.

Path B: own it. A one-time install of the same scope — wired into your phone system, your CRM, your follow-up cadence, running on your accounts. I'll use $9,000 as the build number, because that's a realistic mid-range quote for this scope; below I'll show you what happens when the quote is different. You also pay to run it: model usage and hosting, call it $80/month for this volume. And you budget maintenance, because owned systems need it — say $750 a year for tune-ups when a vendor changes an API or your process shifts. (If anyone selling you a build says maintenance is zero, that's the moment to leave.)

Now the table any investor can rebuild on a napkin:

Read it honestly, because it does not say what a salesman on either side would want it to say.

Year one, renting wins on cash, clearly. $6,000 versus $10,710. If your horizon is a year, or you're not sure the workflow matters yet, renting is the correct trade and I'll say so again below.

At 24 months it's roughly a wash on cash. $12,000 versus $12,420. Anyone who tells you owning "saves you money" over a typical horizon is rounding in their own favor. The cash difference at two years is noise.

The decision is the month-25 line. From month 25 on, the renter pays $500 every month, forever, and holds nothing at any exit. The owner pays about $140 a month all-in and holds an asset — a system that, in the words of the Five Levels post, goes with the company if you sell it tomorrow. Systems and processes are part of what a buyer pays for. A subscription in the vendor's name is not transferable; it isn't yours to sell.

And run the sensitivity, because the build quote is the swing variable: at a $5,000 install the crossover lands around month 14. At $15,000 it's out past month 40, and at $25,000 the subscription wins on cash for more than five years and the terminal-value argument has to carry the entire case — which it might not. The point isn't that owning always wins. The point is that this is a two-minute calculation, the vendor's pricing page and the builder's quote give you every input, and almost nobody at an "AI employee" pitch does it.

Five questions to ask before you sign either contract

You don't have to take my framing. Just ask these, of any vendor and of any builder — me included. They're all verifiable, and how the answers land tells you which side of the table the deal favors.

1. What do I keep if I cancel? Prompts, sequences, conversation history, contact data, phone numbers. Ask for the export format, not a reassurance.
2. What's the all-in 24-month number? Base plan, plus per-seat, plus per-conversation or usage overages, times 24. Compare that to a build quote plus 24 months of running costs, like the table above.
3. Whose accounts does it run on? If the phone number, calendar, and integrations live in the vendor's tenant, that's the depth of the moat you'd have to climb out of later.
4. Can I see the logs? When it says something odd to a seller at 11pm, can you pull the transcript and the reason? (Owned systems get this by construction — it's the entire point of Level 4.)
5. What happens to my workflows if the company is acquired or shuts down? Not an insult — an underwriting question. You'd ask it about a property manager holding your deposits.

A vendor with good answers to all five is a vendor you can rent from with confidence. They exist. This list isn't a trap; it's a filter.

When renting IS the right call

This is the part I'd want to be held to, because I've already written a whole post telling you to buy subscriptions — and nothing here contradicts it. Rent when:

You don't yet know if the workflow matters. Three months of a $500 subscription is a $1,500 experiment. That is the cheapest market research you will ever buy, and it's dramatically better than commissioning a $9,000 build for a workflow you abandon. Rent the experiment. Own what the experiment proves.

The pain is generic. Email triage, meeting notes, generic scheduling — mature product markets at $10–40 a month, shared across thousands of customers, improving weekly on someone else's R&D budget. A custom build cannot compete there and shouldn't try. That's the whole thesis of the buy-it-first series.

Nobody in your shop will own an asset. An owned system needs an owner — someone who notices when it drifts and calls for the tune-up. Even Level 5 needs an owner; I said so in the Five Levels post and it's truer for a small shop. If that person doesn't exist in your operation, a subscription's maintenance-included model is genuinely worth its premium.

Your volume doesn't clear the bar. Twenty leads a month doesn't justify $9,000 of infrastructure. Run the table with your numbers; if the crossover lands in year three or later, rent.

The dividing line, one sentence: rent products, own your process. The moment the thing you're paying for monthly is mostly your configuration — your buy box, your scripts, your follow-up logic, your market knowledge, entered by you, tuned by you — you are renting your own business back. That's the line where the subscription stops being a product and starts being a lease on an asset you already half-built.

What I'd have said in the Q&A

If I'd had the mic on Sunday, I wouldn't have argued against the product. I'd have asked question one — what do I keep if I cancel? — and then suggested everyone open the pricing page and multiply by 24. Some of the room would still have signed up, and for the ones in experiment mode, that's the right answer. The difference is they'd have signed a lease knowing it was a lease.

You already know how to make this decision. You make it every time you tell a tenant-minded friend why you own the duplex you live in, and every time you keep renting the office because tying up capital there is wrong for you. Neither answer is a virtue. The virtue is running the table first.