Five months after closing a modest pre-seed round, a San Francisco startup called Drafted has quietly amassed more than 90,000 user-generated home designs—a respectable number for a platform that began with little more than text prompts and consumer ambitions. But the real story isn't the tally. It's what founder Nicholas Donahue has been building in the margins.
Over recent weeks, communicated primarily through individual LinkedIn posts rather than any coordinated media push, Drafted has rolled out a collection of AI tools that edge it closer to professional territory. Users can now sketch floor plans by hand and watch layouts materialize in real time. They can upload images and convert them directly into Revit models. There's even an early-stage lot automation feature that validates setbacks and garage placement against specific parcels, then spits out views formatted for permitting.
It's an aggressive pivot—or perhaps an expansion—for a company that launched with homeowners in mind. The question is whether Drafted can serve both audiences without losing focus.
The sketch-to-plan interface works the way you'd expect: outline any shape on a digital canvas, and Drafted generates options. A "Remix" button lets users grab an existing design from the platform's public gallery and tweak it rather than starting cold. Practical enough for someone daydreaming about a custom build. Less clear is how many of those 90,000 designs ever made it off a screen.
More interesting, at least from a professional standpoint, is Image2Revit. Built by team member Rio Aguina-Kang, the tool takes an uploaded floor plan image and produces a Revit model with dimensioned walls, doors, windows, and automatic alignment. For architects already embedded in Building Information Modeling workflows, that's a genuine time-saver—assuming the outputs are clean enough to trust.
Drafted now exports DXF and IFC files with properly layered groupings, the sort of thing professionals need for CAD and BIM environments. Mobile browsing arrived recently, too. Small details, but they signal intent.
Donahue has walked this path before, and not always successfully. He launched Atmos through Y Combinator's Summer 2020 batch, raised somewhere in the neighborhood of $20 million, and shut it down earlier this year. Drafted is his second swing at affordable custom home design, this time backed by Convective Capital's Bill Clerico, Stripe's Patrick Collison, Lattice's Jack Altman, and investor Evan Moore, among others.
When the $1.65 million pre-seed round was announced in December 2025, TechCrunch reported that Drafted had six employees and was pricing complete plans between $1,000 and $2,000. The company hasn't updated those figures publicly since. LinkedIn now lists the team as somewhere between two and twelve people, which is either refreshingly transparent or a sign that headcount remains in flux.
The growth metrics, such as they are, come mostly from investor posts and founder updates. Moore noted in December that Drafted had generated 25,000 designs within its first week. By April, Donahue said the platform had crossed 50,000. The homepage counter in late May showed 91,620. Clerico cited 10,000 signups per month in a December post, though these are self-reported numbers rather than audited metrics.
Still, the trajectory is clear: rapid adoption among homeowners and hobbyists, with a smaller but growing cohort of professionals testing the waters.
Drafted isn't operating in a vacuum. Higharc announced sketch-to-3D capabilities in February of last year, targeting homebuilders with automated drafting and estimating integrations. Snaptrude launched an AI collaborator for architects earlier this year, emphasizing site-aware and code-compliant generation. InQI released its property intelligence platform in early May, leaning on a multi-LLM consensus architecture designed to build credibility in AI outputs.
The market is fracturing along predictable lines: consumer tools that prioritize speed and visual appeal on one end, professional platforms promising code compliance and workflow integration on the other. Drafted appears to be straddling both worlds, which is either strategic or precarious depending on execution.
A recent blog post from AI Home Building argued that most AI floor plan generators "miss building physics," pointing to gaps in solar orientation, HVAC placement, and structural logic. Fair criticism. Drafted's lot automation tool suggests an attempt to address site constraints, but jurisdiction-by-jurisdiction code compliance remains unverified. The platform's public gallery skews heavily toward ranch-style layouts and modest square footage—consistent with its affordability positioning, but not exactly serving the high-end custom market.
When Drafted launched, it offered only single-story plans. Multi-story designs were noted as "coming" in December 2025, though the company hasn't publicly confirmed whether that feature has shipped.
The harder question is whether Drafted can scale professional adoption without alienating the consumer base that drove early traction. Image2Revit and IFC exports are table stakes for architecture, engineering, and construction workflows. Lot automation is a step toward permitting, sure. But it's not the same as stamped, permit-ready drawings from a licensed architect or engineer—and no amount of AI sleight-of-hand changes that regulatory reality.
Then again, 90,000 designs in five months suggests there's real appetite for faster, cheaper conceptual work. Whether that appetite translates into sustainable revenue at scale is the problem every startup in this space is trying to solve. Donahue has the scar tissue from Atmos to remind him how quickly things can unravel. This time, perhaps, he's moving more carefully.
