The estimate is the small hinge on which a very large roof swings. Before shingles arrive, before ladders lean against gutters, before a crew clocks in, someone has to turn a jumble of measurements, photographs, prices, codes and judgment into a number a homeowner will sign. It is clerical work with commercial consequences. Send it slowly and the customer may hire somebody else. Get it wrong and the job can lose money with impressive efficiency.
Jahan Khanna likes problems with this sort of leverage. His career has moved through buses, cars, wildfire mitigation and now construction, but the repeating subject is less glamorous than any of those nouns. It is coordination: taking messy information from the physical world and making it useful before impatience wins.
At XBuild, the San Francisco company he co-founded with Rob Moran and civil engineer Sharuk Khanna, that means teaching software to absorb the materials contractors already have. Aerial measurement reports. Jobsite photos. Supplier prices. Local building codes. Insurance formats. The product turns those inputs into an editable estimate and a proposal that can be reviewed, signed and paid without a ceremonial return to the office.
First, a bus that refused to be mysterious
The pattern began at the University of Michigan, where Khanna studied computer science engineering from 2006 to 2009. He was lead programmer on the Magic Bus Project, a transit application designed to answer a question familiar to anyone shivering at a bus stop: how long will I be standing here?
The school project grew into Shepherd Intelligent Systems. Khanna co-founded the company after graduation, and it developed fleet-management and rider-interface systems for public transit, taxis, limousines, school buses and shuttles. The lesson was early and durable. A map is interesting. A map that tells a tired person whether to wait or walk is a product.
Then came a more cinematic ride. In the oft-told origin story for Sidecar, Khanna left a party in San Francisco's Marina District during the winter of 2011 and could not find a cab. A pizza delivery driver took him home; Khanna gave him $20. The exchange suggested a network hiding in ordinary traffic. Cars had empty seats. People needed rides. Smartphones could arrange an introduction.
Khanna joined Sunil Paul and others to build Sidecar, serving as co-founder and chief technology officer. The company was early to the model in which everyday drivers used their own cars to carry passengers. The vocabulary of modern ridesharing had not yet settled. Regulators were suspicious, the apps were young, and the business model still wore the awkward label of a suggested donation.
“Feedback determines about 50 percent of what goes into the app - the other 50 percent are things we think are good ideas.”Jahan Khanna, discussing Sidecar's product process
Khanna's comments from that period sound pleasingly like an engineer in the middle of a real product rather than a founder polishing a legend. Sidecar rebuilt its app from the ground up when the original code no longer fit. The team surveyed users every two weeks. It tested the passenger app across 20 Android devices because fragmentation was, in his word, a nightmare. Lawyers reviewed major features because regulation was part of product development, whether the roadmap liked it or not.
Sidecar also learned the cruel arithmetic of a network business. It raised tens of millions of dollars, expanded across American cities and helped establish product conventions that became familiar across ridesharing. But Uber and Lyft had far more capital. Khanna later told investor and author Andrew Chen that when Sidecar stopped matching the rider and driver bonuses of its larger rival, its markets fell to zero within six weeks.
“It was brutal,” he said. The phrase is more useful than a heroic postmortem. Incentives were not a growth garnish; they held the network together. Sidecar shut down at the end of 2015. General Motors acquired its assets and technology and hired roughly 20 people from the team, including Khanna. He soon left GM for Uber, where he led product for the Vehicle Solutions growth team for nearly four years.
A hard hat for the model
XBuild was founded in 2022, after Khanna had also co-founded Firemaps, a fire-mitigation technology business. With XBuild, the operating environment changed again. There is no two-sided passenger network to subsidize, and no need to persuade a city that a stranger's sedan is public infrastructure. There is instead a stubborn collection of documents and decisions that precedes paid construction work.
The timing matters. Earlier software could store a contractor's information, but a person still had to move that information from photographs and PDFs into fields. Vision and language models can now interpret several kinds of input together. XBuild's bet is that software should perform a meaningful portion of the work, not merely provide a cleaner filing cabinet for it.
Roofing was the opening trade because depth is the difficulty. A credible roof estimate may need an EagleView, GAF QuickMeasure, Hover or Roofr report; pictures from a jobsite; material pricing for the project's ZIP code; and insurance line items that match Xactimate. Building codes can change what belongs beneath a line item. A model that produces a handsome paragraph and misses the flashing is merely an expensive poet.
By January 2026, XBuild and its investors reported more than 15,000 projects completed on the platform, over $250 million in construction value processed and more than 40,000 contractor hours saved. Its end-to-end roofing product could generate an estimate in under 15 minutes. The figures do not prove that every estimate is perfect. They do show a product being used on consequential work.
N47 led a $19 million Series A in January 2026, with Rackhouse Ventures and Andreessen Horowitz participating. XBuild's stated expansion list runs beyond roofing into concrete, landscaping, painting, windows and doors, glass and glazing, insulation, HVAC and plumbing. Each trade brings its own vocabulary, pricing logic and expensive ways to be almost right. The strategy is therefore “deep before wide”: learn one trade's real inputs, then earn the next.
The existential test
In a June 2026 conversation with N47 partner Matthew Cowan, Khanna described the question he asks before building a feature: What is the existential problem being solved? Is there a true risk that the customer's company could go out of business without solving it? Then comes the second test: can the product make the status quo 10 times better?
A two-times improvement can sound excellent and still fail to change behavior, he argued. This is particularly sharp in construction, where a new tool must compete not only with another vendor but with habit, a spreadsheet somebody trusts, a paper form that has survived three software migrations and the colleague who simply knows how the carrier likes it.
“How are we making it 10 times better than the status quo?”Jahan Khanna, N47 Builders interview
This product philosophy gives Khanna's second major founder chapter an intriguing contrast with his first. Sidecar faced a market in which capital could briefly manufacture liquidity in each city. XBuild attaches itself to an existing transaction. The contractor already wants to sell the roof. The homeowner already wants to understand the price. The estimate is the tollbooth between them.
It also clarifies why the company's ambition is larger than faster paperwork without requiring mystical language. The estimate gathers a compact model of the job: what exists, what happened, what materials cost, what the rules demand and what work should follow. If that information stays coherent through proposal, signature, payment and project tracking, it can become a useful record for contractor and customer alike.
What failure leaves behind
Founders are often encouraged to turn failure into a tidy parable. Sidecar resists tidiness. It was early, inventive and influential; it also closed. Competitors adopted pieces of the model and won the market. The company was acquired, but not in the triumphal manner suggested by the word “exit.” Khanna's own recollection preserves the uncomfortable part: a network can disappear much faster than it was built.
What survived was judgment. Khanna had watched a university prototype become operational infrastructure. He had rebuilt an app because the first architecture had become a constraint. He had seen regulation enter the sprint plan and incentives become physics. He had worked inside the larger competitor after losing to it. Those experiences do not guarantee XBuild's outcome, but they explain its concentration on painful work and measurable time.
His public aspiration for XBuild is broad: make it cheaper, faster and easier to build things in the physical world. The phrase could float away in lesser hands. Here it is tethered to a roof, a photograph, a price and a proposal waiting to be signed. Ambition benefits from ballast.
The most revealing line in XBuild's current pitch is not about artificial intelligence. It is the promise that a sales representative can finish the proposal before leaving the driveway. There is a whole biography in that driveway: the bus that was late, the cab that never came, the Sidecar driver trying to find a rider, the contractor trying to turn a visit into a job. Khanna keeps returning to the moment when physical work pauses for information.
Now he would like the pause to last minutes, not days. The clipboard may object. It has had a very good run.