We sold our first ten robots from a folding table at a warehouse expo

The hardest stretch was the second batch of ten units. We had the orders. We didn't have a manufacturing line. We were assembling units in our Hayward space at a rate of about one every five days. Customer #4 from the MODEX batch — a distributor in Cincinnati — emailed me in June 2024 asking why their robot was 38 days late. I had no good answer. I told them honestly that we were a tiny team assembling by hand and we'd ship by week eleven. I offered a 15% refund.

They took the refund and a delayed delivery, and they wrote a one-page letter on company letterhead a month later thanking us for being straight with them. That letter became our most-used sales asset.

§The Anduril offer

The nearest the company came to dying was December 2024. Felicity got a job offer from Anduril at a comp package I literally could not match. She came to my apartment on a Sunday morning and told me. We talked for six hours. I offered her additional founder vesting tied to a 2026 raise. She turned it down. I told her I'd respect whatever she decided.

She stayed. She told me later it wasn't the equity. It was that we'd been honest with the Cincinnati distributor and that, in some hard-to-articulate way, mattered more to her than the money.

The unlock that turned us into an actual company came from a feature request we'd dismissed twice. A produce distributor in Salinas asked whether our robot could integrate with their SKU labeling system, which was an old Honeywell handheld scanner protocol from roughly 2009. Yuri said no. The customer offered to pay $40,000 for the integration. Yuri still said no.

Felicity, who was annoyed at both of us, built it in nine days over Thanksgiving 2024. We charged the customer $40,000. Within sixty days four other produce distributors had signed for the same integration. The Honeywell-compatible variant became 60% of our 2025 revenue.

As of April 2026 we've shipped 184 units, total revenue is approximately $5.2M, and we've raised a $14M Series A from a logistics-focused fund. We use Hetzner for the fleet management API, Tailscale to mesh into customer warehouses for support sessions, and Linear for issues. Gross margins on the latest production run with a Tijuana contract manufacturer are 42%. We still don't run pilots. Customers buy outright or they don't buy.

§From folding table to fleet

The eighteen months after MODEX 2024 were when Stilt stopped being a hardware demo and started being a deployable system. We moved from twelve units in the field to one hundred and seventy. The mechanic was a single product change: we shipped the ROS 2 fleet-coordination layer in November 2024, which let a customer run six Stilt units in the same aisle without manual zoning. 3PL customers — the people running warehouses for Amazon, Walmart, and Target as third parties — went from "interesting demo" to "purchase order" within a quarter once that landed.

We hired our first VP of customer success, Kenji, in early 2025. He'd been a deployment lead at Symbotic and he understood that selling robots to a warehouse manager is 20% selling and 80% making sure the robot does not embarrass them in front of their boss in week one. Our fleet uptime climbed from 88.4% to 97.2% under his playbook. The hardware-sales pattern overlapped almost exactly with what the Detroit industrial-sensor team in this cohort found — that supply-chain credibility from a known city or a known operator is worth more than any product spec.

1. 1
   Pilots are a customer's way of paying you in patience instead of money.

   Skip them. Sell the unit. If the price is wrong, fix the price. Don't trade revenue for "validation."

2. 2
   Sub-$30k unlocks the corporate card.

   Anything above that pulls procurement into the deal. Procurement turns a one-week sale into a six-month sale. Price for the buying motion you want.

3. 3
   The booth at the unsexy expo beats the venture announcement.

   MODEX and ProMat are where actual warehouse buyers are. Y Combinator's batch demo day is not.

4. 4
   Honesty about your delay is a sales asset.

   The Cincinnati letter is still the third paragraph in our sales deck.

5. 5
   The integration you don't want to build is the one your customer will pay $40k for.

   Listen to the second customer who asks. That's the signal.

What this looked like in practice.

The thing nobody tells you about *selling hardware at trade shows* is that the people who buy at the booth are not the people who sign the PO. The booth gets you the email; the PO comes from a Tuesday-afternoon call six weeks later with someone who never came to the show.

The thing nobody tells you about *robot demos* is that you do not demo the robot — you demo the dashboard that proves the robot did not crash overnight. Warehouse operators do not care about your kinematics; they care about whether they will get a 3am phone call.

The thing nobody tells you about *fleet management* is that two robots are easier than one, six are harder than two, and twenty are easier than six. The complexity is non-monotonic. We built the ROS 2 layer assuming linear scaling and were wrong by an order of magnitude in the middle range.

The thing nobody tells you about *3PL customers* is that they will pay more per unit than the brands they serve, because their margins are thinner and a single missed peak season costs them a contract worth ten times the robot.

The thing nobody tells you about *hiring a deployment lead before a salesperson* is that uptime is the salesperson. Our second-year revenue was almost entirely expansion within accounts Kenji had stabilized. A unhappy first deployment costs you the next ten in the same logo.