Drafted, a Y Combinator backed model from the 2026 cohort, turns a rough footprint sketch into a complete floor plan and exterior in seconds. Buildability, code, and permits are where the output stops.
From a rough sketch of a home's footprint, Drafted returns a complete floor plan and matching exterior in seconds. In a single month, 120,000 users generated 325,000 home designs on the platform, according to founder Nick in the company's Y Combinator Launch post. That usage figure is the most concrete signal yet of where residential pre-design is heading, and the clearest way to see what today's generative architecture tools can, and cannot, do.
Drafted, part of Y Combinator's 2026 batch, is built around a deliberately narrow idea. A user types or draws a few structured constraints: target square footage, the shape of the home's footprint on the lot, lot boundaries, where specific rooms should sit, and how spaces should relate to one another. The model then produces a two-dimensional floor plan and a matching exterior elevation, exportable as CAD and PDF files, as described on the company's site and in a public demo video. The founder's framing of the inputs as "constraints" is the technical heart of the system. It is what makes the output feel like a designed home rather than a generic rendering, and it is the part of the product most likely to generalize beyond Drafted itself.
The hard part begins after the file is exported. A floor plan that fits a footprint and respects user-specified adjacencies is not, by itself, a buildable document. It does not show load-bearing walls, foundation requirements, structural spans, mechanical routing, egress dimensions, fire separation, or local zoning rules. None of that is in the model's input set, and none of it appears in the output. A two-dimensional plan also cannot resolve the questions that determine whether a home actually gets built: soil conditions, utility connections, setback compliance, and the long list of code checks a building department applies before issuing a permit.
That gap is not a flaw specific to Drafted. It is the boundary every generative residential design tool currently sits behind, and it is the line the product is honest about. The company describes its output as pre-design, not permit-ready drawings. Users who want construction documents still need a licensed architect or engineer, and most jurisdictions still require stamped drawings for any new home. The founder's claim that "the vast majority of homes are built without direct architectural involvement" is true, and it is also a reason for caution. The homes that skip architects tend to be the ones where production builders replicate a small set of stock plans, not the ones where a generative tool is producing a novel layout for a specific lot.
The 325,000-designs-in-a-month number is best read as a stress test of pre-design demand, not as a verdict on the future of architecture. It shows that a large number of people want to see what a home could look like on their lot, fast, before committing to the cost of a full design engagement. Custom home design typically costs $10,000–$50,000 or more and takes months — and the first weeks of that process are usually spent doing exactly what Drafted does in seconds: sketching footprint shapes, testing room adjacencies, and getting a rough sense of what fits. For that stage of the workflow, the tool has genuine economic value. For everything downstream, the workflow has not changed: an architect, a builder, and a building department still have the final word.
The economic question the headline asks — what are 325,000 generated designs actually worth — resolves to a simple test. Today the model accepts size, shape, and adjacency as constraints. The output is worth something to the extent that those are the only inputs that matter for a given buyer's decision. A homeowner refining a concept before meeting an architect will find real value in seconds of iteration. A developer expecting a permit submission will find the same output is not yet worth anything at all. The constraint set is the mechanism that determines where on that spectrum any given user lands. What to watch next is whether it grows — whether future versions ingest zoning envelopes, climate data, structural systems, and code constraints — because that is the moment the pre-design output and the permit-ready drawing start to converge, and the economics of custom home design begin to shift in earnest.