The Legend of Steve, and the quiet machine Rohan Chopra built next
An early DoorDash engineer watched a company free one overworked dispatcher instead of firing him. A decade later, that idea is a $38M-backed startup teaching software to do the office work nobody ever bothered to automate.
In the earliest days of DoorDash, the whole delivery network ran on one person. His name was Steve. An order would land, and Steve would look at who was driving nearby, pick the closest one, and text them to go grab it. No routing algorithm. No dispatch system. A guy, a phone, and a growing pile of orders.
Rohan Chopra was one of the engineers who joined DoorDash around that time, when the company was still a couple dozen people and the "logistics engine" was mostly Steve's thumbs. Chopra built the software that took the brokering off Steve's plate - the code that decided which driver got which order, at the scale of millions of deliveries. And here is the part he keeps coming back to: the company did not fire Steve. It moved him to harder problems that only a human could solve.
That story is now told, half as folklore, inside the startup Chopra runs. People there call it the Legend of Steve. It is also, more or less, the entire thesis of the company.
01 / The SetupEight years learning where software stops
Chopra studied computer science at Stanford and landed at DoorDash as one of its first twenty employees. He stayed roughly eight years, which in startup time is a couple of geological eras. He left as a senior engineering leader running the Dasher and logistics group - the routing, the supply-and-demand balancing, the unglamorous machinery that has to work every single time or the food gets cold and the customer never comes back.
Building that kind of system teaches a specific lesson: reliability is the product. A demo that works once is worthless if it fails on run number four hundred thousand. That instinct shows up later, in how he talks about AI - not as a party trick, but as something that has to hold up under a real operating load.
It also gave him a front-row seat to a pattern he could not un-see. Across companies, smart and well-paid people were spending two to three hours a day on work no engineer had ever gotten around to automating. Clicking buttons. Pulling the same report. Reconciling invoices. Copying a number from one system into another. The tasks were too fiddly for off-the-shelf software and too small to justify a dedicated engineering team. So they just sat there, on human shoulders, year after year.
02 / The BuildThree childhood friends and a different kind of AI
In 2024, Chopra started Convey with two childhood friends, Diego Canales and Will Harvey. It is a nice detail that the co-founders go back that far; two of them had already built and sold a company to project44 before regrouping for this one. When Andreessen Horowitz wrote up its investment, it led with the team, not the market: anyone who has worked with one of Convey's three founders, the firm said, will tell you not to pass up the chance to work with them again.
The product is deceptively simple to describe. Convey lets a non-technical person build an AI "teammate" that does a real operational job - processing invoices, reconciling accounts, pulling campaign reports, wrangling ad assets. You do not write code. You onboard it the way you would onboard a new hire: share your screen, walk through the process once, and let it watch. In about three hours, it can start doing the work.
The technical choice underneath is where Chopra's DoorDash reflexes show. Most enterprise AI tools run a fresh prompt every time they act, which means they can be brilliant on Tuesday and baffling on Wednesday. Convey's agents instead compile what they learn into real, versioned, testable programs - software you can inspect, roll back, and trust across hundreds of thousands of runs. It is the difference between a clever improviser and an employee who actually remembers how you like things done.
03 / The ProofA million hours, before the round even closed
By the time the funding was announced in June 2026, Convey's teammates had already logged more than a million hours of real work. The customer list reads like a cross-section of companies that cannot afford flaky software: NBCUniversal, Samsara, TelevisaUnivision, Unity, Faire, ChargePoint. At Faire, invoice processing that had eaten hundreds of manual hours got handed to a teammate. One streaming customer clawed back more than 23,000 hours a year across reporting and ad operations. A logistics firm, Savoya, credited the shift with a 40 percent year-over-year jump in EBITDA and roughly ten thousand reclaimed hours.
The human details are the ones that stick, though. At Ewing Outdoor Supply, a talent manager got promoted to run automation full time. At Northwest Meat Company, a nightly task nobody wanted to staff quietly started running itself. Neither story is about a job disappearing. Both are about the Legend of Steve, playing out again - someone freed from the boring part and pointed at something that matters more.
04 / The Money$38M, and a bet on the team
The Series A was $38 million, led by Andreessen Horowitz with partners Joe Schmidt and Olivia Moore, and joined by Khosla Ventures and Pear VC. Schmidt took a board seat. Khosla's Samir Kaul described Chopra in terms of grit and creativity - the two traits that tend to matter when a category is crowded and everyone is racing to say the word "agent" first.
What is striking about Chopra's framing is how little of it is about the technology itself. He talks about operational drag - the friction that slows a company down without ever showing up cleanly on a spreadsheet. His argument is that the companies winning right now are the ones removing that drag, scaling through a digital workforce whose impact you can actually measure in hours returned. It is a curiously old-fashioned pitch for an AI company: not "look what the model can do," but "look what your people could do instead."
05 / The PointGive it back so they own it
There is one more line of Chopra's worth sitting with, because it is a strange thing for an AI founder to say out loud. The goal, he explains, is to parachute in, help a team stand up its AI workforce, and then hand the controls back so the customer owns it. No black box. No permanent dependency. In an industry that mostly sells lock-in, that is close to heresy.
It also completes the circle back to Steve. The whole point of taking the brokering off Steve's plate was never to make Steve disappear - it was to give him better problems and let him own the outcome. Fifteen-odd years and a million automated hours later, Chopra is running the same play, just for everyone else's Steves: the operators quietly holding companies together with spreadsheets and patience, waiting for someone to finally write them some software.