Money & property · Sep 2026 · with 
Muse
Concluded
Killing a car search with math
In September I wanted a bigger Tesla: a used 2024–2026 Model X, under 30,000 miles, black, gray or blue. I set M, my agent, to scan listings every few days and report only new matches. The search ran for about three weeks. It ended not with a purchase but with a spreadsheet — and the spreadsheet is the reason this write-up exists.
The ground rules
- No prior repairs or accident history. Non-negotiable, set before the first scan.
- No dealer contact. M researches and reports; nobody talks to a seller until my wife and I say go.
- Only new matches. A seen-list meant every report showed cars I hadn't already looked at — no recycled listings dressed up as news.
The Marietta car
The strongest early candidate was a 2024 in stealth gray: 17,725 miles, $69,997, no dealer fees. Its listing page even showed the reassuring structured badges — clean title, zero accidents reported. But buried in the dealer's own written description was a disclosure that the car had a manufacturer buyback / lemon-law history.
That ended it in one breath. The lesson is worth more than the car: the tidy history summary and the seller's prose are different documents, and the prose is where the truth was. An agent that reads the whole listing — including the boring paragraphs — catches what the badges smooth over. A second car was excluded the same week for an accident record. The rule held because it was written down before emotions got a vote.
The five-year model
While the scan ran, M built the comparison that actually mattered: not sticker prices, but five-year total cost of ownership across four options — keep our 2021 Model Y (~56k miles), buy a new Model Y, buy the larger new Model Y L, or buy the used Model X at $72,190. Researched inputs, all checkable: California insurance averages by model, home charging at off-peak rates, tire sets and realistic tire life, remaining battery warranty, depreciation curves, a maintenance reserve for the X's falcon-wing doors and air suspension, and no federal EV credit anywhere (the credits expired for vehicles acquired after September 30, 2025 — a fact that quietly changed a lot of used-EV math that year). California also gives no sales-tax credit for a trade-in, which the model handles correctly and many back-of-envelope versions don't.
| Cost item (5 years) | Keep 2021 Y | New Model Y | Model Y L | Used Model X |
|---|---|---|---|---|
| Net cash outlay (after trade-in) | $0 | $31,223 | $45,552 | $56,158 |
| Insurance | $20,575 | $20,575 | $20,575 | $26,920 |
| Charging | $3,726 | $3,726 | $3,726 | $4,830 |
| Tires | $3,200 | $3,200 | $3,200 | $3,600 |
| Maintenance & reserves | $1,000 | $1,000 | $1,000 | $2,500 |
| Less: value at year 5 | −$13,000 | −$20,000 | −$25,000 | −$26,000 |
| Total cost of ownership | $15,501 | $39,724 | $49,053 | $68,008 |
Planning estimates from September 2026, 12,000 miles/year, trade-in variants shown. One toggle matters: Tesla had stopped making basic Autopilot standard on new base trims, so lane-keeping on a new car can require the $99/month FSD subscription — $5,940 over five years wherever the toggle is on. Our current car and the used X keep their Autopilot free.
Artifact: the actual workbook — Tesla 5-Year Ownership Analysis (.xlsx) — assumptions on one sheet (edit the yellow cells), comparison on the other, every total a formula. Personal details scrubbed; there weren't many to begin with.
The verdict
The used X wasn't a little more expensive. It was the costliest option by roughly $28,000 over five years versus a new Model Y — before counting that our current car's steep depreciation has already happened, so keeping it is riding the flat part of the curve. On September 29 I killed the search entirely: scan deleted, tracked item closed, no "just keep an eye out" afterlife. We're leaning toward a new Model Y or Y L in late 2026 or early 2027 instead, and no pricing watch was set — when we're ready, we'll run the model again with fresh numbers.
The agent's most valuable output all month was not a car. It was a defensible reason to stop shopping.
Result
Three weeks of scanning, two cars disqualified on evidence, one spreadsheet, zero dollars spent, zero dealers contacted. This is a companion piece to entry 20 of AI Lab Work — the car comparison is one of the jobs it mentions.