Starship Technologies
100K+ deliveries / week
Live on ~50 US college campuses. Six-wheeled, 20 lb payload, works in rain. Proves sidewalk autonomy is a solved problem for <1mi trips.
~/racct
/
memos
/
autonomous_only.md
$ cat autonomous_only.md_
HUMANS ARE v0. FLEETS ARE v1.
The winner of local delivery in 2028 has a fleet, not drivers. This memo lays out the leapfrog, the wedge, and the case to the team.
100K+
Starship deliveries / week on campuses
1M+
Zipline autonomous deliveries to date
2028
The year the moat closes
0
Human drivers in the target end-state
// 01
the leapfrog
Route by distance and payload. Match the right autonomous form-factor to the trip. Humans don't enter the graph.
tier
distance
form-factor
payload
unit econ
T1
< 1 mi
sidewalk bots
< 20 lb
low CAPEX / high density
T2
1 – 5 mi
drones
< 8 lb
air = no traffic tax
T3
5+ mi
autonomous vans
< 500 lb
multi-stop batching
// proof points already in the wild
Zipline
1M+ autonomous flights
Rwanda national blood supply since 2016. Now running for Walmart in Arkansas & Sweetgreen ops. P2 platform hovers and lowers a package on a tether. This is the 1-5 mi solution.
Coco Robotics
500K+ LA deliveries
Restaurant-grade sidewalk bots across LA, Austin, Miami, Helsinki. Contracted with DoorDash + Uber. Dense-urban validation, not just campus.
Nuro / Cartken
L4 autonomy on public roads
Nuro's driverless vans have racked up hundreds of thousands of grocery + retail miles for Kroger, Domino's, 7-Eleven. Cartken adds mid-tier robots for Uber Eats. The T3 layer exists.
// racct's edge
Every operator above is vertical: one form-factor, one category. racct is category-agnostic routing — a resale bike frame, a takeout burrito, a returned Amazon box, and a prescription bottle all enter the same request layer and get dispatched to the right tier by distance and payload. We don't build the fleet. We're the exchange that clears it.
// 02
the path
Pick one campus/urban wedge in a PAVE Act state. Rent fleets. Ship local resale first. Compound from there.
// wedge selection
geography
- > Austin, TX (PAVE Act; Starship + Coco + Avride live)
- > Columbus, OH (PAVE Act; dense student pop; Nuro history)
- > Tempe, AZ (PAVE Act; Waymo halo; ASU campus)
All three ship autonomous rigs on public sidewalks and roads without per-city fights. Regulatory tailwind, not headwind.
category
- > Local resale first (couches, bikes, monitors)
- > Facebook Marketplace's biggest unsolved pain: who moves it
- > Buyer + seller both want a bot, not a stranger's Civic
Resale is the perfect wedge: high-margin trips, low SLA sensitivity, and every completed handoff is content for the feed.
// operating model — rent, don't own
We do not buy robots. We sign fleet-as-a-service contracts and route across them:
Serve Robotics
Coco
Cartken
Starship
Nuro
Zipline
Avride
CAPEX stays on their balance sheet. We ship a routing layer, a marketplace, and a brand. The graph of vendors is our moat, not the wheels.
// 3-phase rollout
PHASE 01
months 0-6
wedge // one city, one bot vendor
- > Austin. Sign Coco or Avride. T1 only.
- > Local resale + campus errands.
- > Target: 1,000 deliveries/week, <1 mi.
- > Prove SLA, price point, and repeat rate.
PHASE 02
months 6-18
stack // add tiers, add cities
- > Layer T2 (Zipline / drone partner) for 1-5 mi.
- > Layer T3 (Nuro / Cartken van) for 5+ mi.
- > Open Columbus + Tempe. Same playbook.
- > Expand categories: retail returns, groceries.
PHASE 03
months 18-36
clear // become the exchange
- > 15+ metros, 5+ fleet vendors.
- > Routing engine picks vendor per trip in real time.
- > Marketplaces, retailers, and brands plug in via API.
- > racct = the demand aggregator for autonomous delivery.
// 03
the pitch to the team
Read this before the next standup. If we're honest about what we're building, some of what's on the roadmap this week is off it.
01
humans are v0.
Every human-driver marketplace ever built has hit the same wall: acquisition cost climbs, retention rots, insurance eats margin, and the driver is also the churn. We are not going to be the one company that finally makes gig work profitable. We're going to be the company that skipped the era.
02
every hour on driver retention is an hour not on the moat.
Onboarding funnels, background checks, driver support, referral bounties, insurance policy, 1099 tax UX — none of it compounds. The routing engine, the vendor graph, the marketplace flywheel, the API — all of it compounds. Every eng-week we spend making humans work is a week Uber, DoorDash, and Amazon are already ten years ahead on.
03
the winner of local delivery in 2028 has a fleet, not drivers.
By 2028 the cost curve on autonomous delivery crosses the cost curve on human delivery in every dense metro. The incumbent playbook is to bolt bots on top of a driver network. Ours is to start on the other side of the curve and never build the driver network at all. That is the whole bet.
path a // more of the same
- hire drivers
- fight for supply
- race Uber and DoorDash on their turf
- margin capped by 1099 economics
- indistinguishable from every marketplace since 2013
path b // the leapfrog
- rent fleets, don't own them
- route by distance + payload
- own the demand + the exchange
- margin expands as fleets get cheaper
- we are the only company on this path
$ racct --mode autonomous --drivers 0
> committing to the leapfrog.
> deprecating v0.
> the moat starts monday.