What the business is
Generates buildable residential floor plans, 3D layouts and renders from lot size, square footage and plain-language description, charging $1,000–2,000 for a complete construction-ready plan.
The bet
That a generative model trained on buildable, permitted plans can replace staff designers entirely and sell custom home plans at $1,000–2,000 instead of $10,000–50,000.
Starting capital
$17.5M total — $1.65M pre-seed at a $35M post-money valuation (Dec 2025), then a $16M seed round led by Buckley Ventures (May 2026)
How it started
Donahue grew up in a homebuilding family and dropped out of NC State to fix why custom home design cost a fortune. His first company Atmos went through YC, raised ~$20M from Khosla Ventures and Sam Altman, and grew to 40 people and $7M revenue — but it was software fronted by staff designers, what Donahue calls a glamorized architecture firm. When the Fed jacked up rates, clients who had spent months designing could no longer afford to build, and he shut Atmos down in early 2025.
What happened
Nine months later he started Drafted with six employees, four from Atmos, and no designers at all. Its own model, trained on real plans from homes that were built and passed permitting, costs about 0.2 cents per floor plan versus 13 cents for general-purpose AI. Bill Clerico spent two weeks telling him 'please take our money' and led a $1.65M pre-seed at $35M post alongside Patrick Collison, Jack Altman and Warriors player Moses Moody. By Dec 2025 Drafted saw about 1,000 daily users; in May 2026 BI reported a $16M seed round; the past month brought 120k users and 325k+ generated plans. On HN's Launch, users still caught hard errors — cars parked perpendicular to the garage door, a bedroom with no closet — a reminder that buildability is the model's real exam.
How it ended up
Still scaling. Nine-person team in YC's Spring 2026 batch, multi-story and lot-specific features on the roadmap, backed by Samsung NEXT, Ben Silbermann and Buckley Ventures.
What has to be true
Vertical data and unit economics: training on plans that were actually built and permitted makes the output buildable at 0.2 cents per plan, two orders of magnitude below general-purpose AI.
The demand was already validated by the founder's dead company: Atmos designed $200M worth of houses before rates killed it, and Drafted's traffic arrived with zero paid acquisition.
Second-time founder information advantage: four of six original employees, the angel network, and the training corpus all carried over from Atmos.
The bet can still fail on both flagged fronts: most Americans may not want custom homes, and generated plans still miss functional details a permitted drawing must get right.
What can be applied
Validate demand with a service, then subtract the humans: Atmos proved the demand but died carrying designer payroll. The rebuild kept the demand, deleted the cost structure.
Aftermath
As of September 2026 Drafted is a nine-person team in YC's Spring 2026 batch that has raised $17.5M across two rounds. Its YC listing reports 120k site users and 325k+ floor plans generated in the past month, driven by social sharing and word of mouth. The product still covers single-story homes only, with multi-story and lot-specific design on the roadmap; users include homebuyers (over a third of visitors), architects, builders and developers.
FOLLOW THE EVIDENCE
The sources
- This founder just landed funding for a second go at the same problem: affordable custom home design techcrunch.com
- Drafted Secures $16M in Funding to Reshape Home Design With AI businessinsider.com
- Drafted — Y Combinator company profile ycombinator.com
- Launch HN: Drafted (YC P26) – Models for residential architecture news.ycombinator.com