turobot

an experiment at johnnys.ai

A real business,
learning to run itself.

One Cybertruck, rented on Turo. Machines built the case for it, watch the market for it, and increasingly propose what it should do next. This line is its scoreboard — every cash dollar and the modeled tax effect that made the business viable.

$0.00

modeled business position

modeled position cashflow modeled tax value† booked revenue
vehicle Cybertruck Model X Model Y

drag across the chart to relive it

Every step is a dated ledger event: the down payment, each month's $1,311.49 loan payment at 0.00% APR, each platform-recorded Turo payment net of fees, and documented upkeep. The fleet-activity strip shows when each vehicle was away; select a window to see its payment cadence. When a real rental begins, its committed host payout enters the bold modeled line immediately; the green field then shrinks as Turo pays in installments and the dotted cashflow catches up. The gold field separately carries the modeled 2025 tax value until the following Apr 15 filing day.† Where the lines overlap, the modeled line stays on top. No smoothing or cherry-picked start date.

–

modeled business position

–

Turo payments in

–

cash out

–

tax value realized†

3 trips completed · 7 platform payment events · 31 days on rent · $172 payout per rented day

01 · THE EXPERIMENT

What is this?

This business exists because language models argued it into existence. The tax case for the truck, the read on rental demand, the economics of every handoff — machine-built first, human-signed after. In November 2025 the household took delivery of a Cybertruck; by New Year's Eve it had earned its first real revenue.

A word of honesty about the word agentic: this is not an autonomous system. Today the machines watch the market, run the numbers, and draft the decisions — and I still make them. Every dollar on this page moved because a human said yes. The experiment is to move that boundary deliberately — pricing first, then fleet — and to keep publishing this page, whichever way the line goes.

DECISION AUTONOMY

  1. ADVISEDmodels argue, human acts
  2. ANALYZEDsystem watches the market
  3. PROPOSEDsystem drafts the decision
  4. DECIDEDsystem commits real money

02 · THE DECISION LOG

The decisions so far

  1. 2025-09-30

    Acquire 2026 Cybertruck AWD

    human decidedagent argued

    The case was machine-built: 100% bonus depreciation clears the §280F cap only because the truck's GVWR does, and the scraped market data said demand was real.

  2. 2025-11-30

    List on Turo

    human decided

    Truck placed in service; first booking arrived within a month.

  3. 2026-05

    Doctrine: prefer longer rentals, ≥$200 contribution per turn

    human ratifiedagent analyzed

    Every handoff costs real hours and washes. The system priced the turn itself.

  4. 2026-07-25

    Accept 24-day rental

    human decideddoctrine-aligned

    Longest trip yet; best payout yet ($3,895.74). The doctrine worked.

  5. open

    Model Y acquisition — availability + FSD backup

    agent proposedpending

    The fleet simulator says aggregate availability is the binding constraint. The human has not yet signed.

  6. next

    Dynamic repricing

    target: agent decides

    The first decision class slated to cross the line — the machine sets the nightly price, and money moves.

03 · THE REGISTER

Every dollar, on the record

DateEventAmountRunning

Provenance — contract installment-sale contract · platform Turo transaction export · receipt invoice on file · estimate reported, receipt unresolved · tax tax value, realized†