Robin Li spent eight years at Arup designing the bones of buildings that other people would get to name. Steel moment frames for towers in downtown Los Angeles, wood-and-concrete hybrids for mid-rise housing in the East Bay, the invisible architecture that keeps the visible architecture standing. In a typical residential project, Li and his colleagues would receive an architect's design, run the structural calculations by hand and in software, and return a single set of beam sizes, column locations, and connection details that satisfied code requirements while staying within the budget. One option. Maybe two if the client pushed back on cost.
Now Li is a cofounder of Genia, an LA-based startup that raised $3 million in seed funding from Pi Labs, Amplify, Boost VC, Suffolk Technologies, and Scale AI executives to build something that would have made his former Arup colleagues deeply uncomfortable: a generative AI system that produces up to five structurally validated design options from a single set of architectural drawings. Not conceptual sketches or rough parametric studies, but actual beam dimensions, coordinates, material specifications, and load paths, each option optimized for a different priority. Cost. Stability. Sustainability. Feasibility. All of them compliant with the applicable building code.
Five options where there used to be one. That multiplication sounds like pure upside until you watch what happens on a job site when someone tries to build Option Three.
What Genia Actually Does
Upload a PDF of architectural drawings or an AutoCAD file. Genia's system parses the geometry, identifies the structural requirements, and runs physics-based generative algorithms that explore the solution space for configurations satisfying the relevant building code. It returns up to five options, each presented with beam dimensions, material quantities, estimated costs, and stability metrics. Genia claims 75% reduction in manual calculation work and up to 20% reduction in material use across the generated options, though neither figure has been independently audited.
This is not a chatbot that summarizes engineering principles. Genia's founding team emphasizes "pixel-level accuracy," meaning the output includes specific coordinates and cross-sections that a detailer could theoretically take straight into shop drawings. They have partnered with Weyerhaeuser ForteWeb and Simpson Strong-Tie for material databases, and signed a deal with Suffolk Construction, one of North America's largest contractors, to apply the tool on adaptive reuse projects where existing structural conditions add complexity that makes manual iteration especially painful.
Dr. Murat Melek, a licensed SE at Suffolk, described the system as combining "AI with code-based checks in a truly innovative way." That endorsement carries weight because Suffolk stakes its reputation on structural performance, and licensed engineers do not publicly praise tools that produce unreliable calculations when their license is on the line.
The Cost-Per-Option Math
Residential structural engineering runs $500 to $3,000 for basic design plans and $2,000 to $20,000 for full new-home structural packages, depending on complexity and location, according to aggregated data from Angi, HomeGuide, and Fixr as of 2026. Hourly rates for licensed structural engineers average $150, ranging from $100 to $250 depending on market. A typical residential project takes four to eight weeks from architectural drawings to stamped structural plans.
At those rates, the cost per explored design option under the traditional model runs between $5,000 and $15,000 per option, because exploring a meaningfully different structural approach requires the engineer to essentially restart the calculation process with new assumptions. Most residential projects therefore explore one option, occasionally two.
Genia generates five options in what the company describes as roughly eleven minutes of compute time. Even at an aggressive SaaS subscription estimate of $2,000 to $6,000 per month for a small firm, the cost per option drops to approximately $200 to $600. That is a 90 to 96 percent reduction in cost per explored structural alternative.
Compelling? Absolutely. Complete? No.
The PE Stamp Still Costs What It Costs
Every structural design submitted for a building permit in the United States requires the seal and signature of a licensed Professional Engineer. California Business and Professions Code Section 6732 prohibits unlicensed individuals from using PE titles or seals. Texas Board Rules Sections 137.33 and 137.77 require firm registration on all engineering documents. New York mandates professional seals with specific procedures for electronic signatures.
AI does not change this requirement, cannot change it, and should not change it.
A PE reviewing Genia's output must independently verify the calculations, check code compliance against local amendments, and accept professional liability for the stamped drawings. If the PE reviews five AI-generated options instead of one, the review fees scale accordingly. At $150 to $250 per hour with two to four hours per option, five options cost $1,500 to $5,000 in PE review alone, eating into the theoretical savings from faster generation.
Net savings remain real, but they are not the 90% headline figure. For a $10,000 residential structural package, AI-assisted design with full PE review of five options might cost $4,500 to $8,000 total, depending on how much of the traditional calculation work the PE can offload to the tool versus how much they need to recheck from scratch because they do not yet trust the system's output on their specific building types and local code interpretations.
MIT Found Something Interesting About the Middle Ground
Researchers at MIT developed what they call "Human-Informed Topology Optimization," a hybrid process that pauses automated structural optimization at key decision points for engineer evaluation before continuing. The results outperformed both fully automated and fully manual approaches.
This matters because it suggests the optimal deployment of AI structural tools is not replacement but interruption. Let the AI explore the vast solution space that humans would never reach on their own, then let the human catch what the AI cannot see: the context, the constructability constraints, the aesthetic judgment, the knowledge that the building inspector in Santa Clara County has strong opinions about moment frame connections that are not written in any code but will absolutely delay your permit if you ignore them.
Genia's multi-option approach implicitly acknowledges this dynamic. Five options is an invitation for human judgment, not a replacement for it. The architect and engineer review the five, apply their own experience and constraints, and arrive at a final design that neither the human alone nor the AI alone would have proposed. That is a genuinely different design process than what exists today in residential construction, where most homes are built from a single structural option that nobody questioned because questioning it would have cost another $8,000 and six weeks.
Constructability Is Where the Theory Breaks
A beam that satisfies every load case in the building code and minimizes material use by 20% is not automatically a beam your framing crew can install. Experienced structural engineers optimize for things that never appear in calculations: the joist hanger that fits the available lumber dimensions at your local supplier, the connection detail your ironworker can execute without custom fabrication, the column placement that avoids the plumbing stack the mechanical engineer has already routed through that wall cavity.
These constraints are invisible to physics-based optimization. They live in the accumulated judgment of engineers who have watched framers struggle with clever designs, who have gotten calls from the field at 6 AM asking why the specified W12x26 beam does not fit through the hallway opening and whether they can substitute a W10x33 that does. An AI that has never stood in a half-framed house watching a crane operator try to swing a beam through a 36-inch gap between the existing structure and the neighbor's property line will generate structurally valid options that are, in the words of one structural engineer I spoke with off the record, "physics-valid but construction-stupid."
Genia's partnership with Weyerhaeuser and Simpson Strong-Tie addresses the material database gap. Their Suffolk deal exposes the system to real construction feedback. But Genia is pre-revenue or early revenue, and no published case study documents what happened when their five options met an actual framing crew.
Who Should Use This Now
If you are an architect or structural engineer working on custom residential projects in the $500,000 to $5,000,000 construction cost range, AI structural tools represent a genuine expansion of your design vocabulary. The ability to explore five structural configurations instead of one changes the conversation between architect and engineer from "can we do this?" to "which of these five approaches gives us the clearest path to what we actually want the building to feel like?" That is a different question, and it produces different buildings.
If you are a homeowner building custom, ask your structural engineer whether they are evaluating generative AI tools. Not to pressure them into adoption, but because the conversation reveals how they think about optimization versus constructability. An engineer who says "I looked at Genia and I think it is promising but I do not trust it yet on wood-frame residential because the connection detailing needs work" is telling you they understand both the opportunity and the risk. An engineer who says "AI cannot replace forty years of experience" is telling you they have not looked at it. An engineer who says "we use AI for everything now" is telling you they may be stamping output they have not fully reviewed. Pay attention.
For production builders running the same three or four floor plans across hundreds of units, the value proposition is different and possibly stronger. Optimizing the structural design of a plan that will be built 200 times means every pound of steel saved or every simplified connection detail multiplies across the entire production run. A 15% material reduction on a floor plan built 200 times is not 15% savings on one house. It is 30 houses' worth of material across the portfolio.
Limitations
Genia's 75% calculation reduction and 20% material savings are vendor claims. No third-party engineering firm or academic institution has independently verified these figures on a controlled set of projects. Suffolk Construction's partnership has been announced but no published project outcomes document actual performance, cost savings, or building department acceptance of AI-assisted structural submissions. The cost-per-option calculation in this article estimates Genia's subscription pricing from comparable SaaS tools because Genia has not published its pricing publicly. PE review time estimates assume the reviewing engineer treats AI output with the same verification rigor as a junior engineer's work, which may understate or overstate actual review time depending on the PE's familiarity with the tool. No data exists on building department acceptance rates for AI-generated structural calculations in any US jurisdiction. Residential-specific performance data is thin because most AI structural tools, including Genia, have focused their early deployments on commercial and multifamily projects where the calculation complexity and potential savings per project are larger. The liability landscape for PE-stamped AI-assisted structural designs is legally untested, with no precedent cases establishing how courts will apportion responsibility when an AI-generated structural element contributes to a failure.