Governments around the world are working hard to build more houses, offices, factories, and data centers faster. There are many bottlenecks and impediments to these goals, and code compliance is one of them. Digital Building Information Models (BIM) are widely used, particularly in the UK and the US and provide some ability to identify many problems early in the design, engineering, and construction processes.
It would be ideal if architects, engineers, builders, and regulators could automatically run a regulatory check whenever they make a change or sign off on a new version. But compliance is a much harder problem to check or test than, say, spelling or grammar. It's probably even harder than compliance automation for software code and services, which is still a very hard problem above and beyond testing for defects, security, and performance issues.
Maor Greenberg, co-founder and CEO of Spacial, explains:
Automating code compliance is hard because building codes aren't necessarily digital-friendly. They are legal documents written in ambiguous, jurisdiction-specific language. Every city, county, or state might enforce a slightly different interpretation of similar rules. Standards like IFC and newer BIM interoperability have helped improve model structure, but they don't solve the semantic complexity or local nuance of interpreting real-world code enforcement.
Spacial recently raised $10 million for a novel approach to automated code compliance that combines AI reviews with a collaborative process that streamlines analysis, edits and reviews with structural, mechanical, electrical, and plumbing (MEP) engineers. This allows architects and builders to move from initial design to stamped plans in 7-10 days, compared to the weeks or months often required by manual processes. Spacial says this can also shave about 25% off approval time by coordinating structural, MEP, and energy reviews into a single process.
Several vendors, such as Autodesk and Solibri, have supported basic compliance automation capabilities for BIM models for years. Others, like UpCodes, have developed natural language processing to support AI research assistants of relevant code libraries. This can be tricky, since there can be dozens of applicable codes that may need investigation as part of a project. Aside from the physical layout, there can be fire, accessibility, green building, performance, and safety aspects to comply with.
Spacial is bridging the gap between these tools by linking sometimes ambiguous regulatory semantics to building plans. This approach has also helped them reimagine architectural and engineering processes. Greenberg says:
The real shift is mindset. We're not just digitizing drawings anymore. We're digitizing the creative process, and that requires combining engineering logic, legal interpretation, and planning workflows.
Greenberg argues that the legacy solutions are rule-checking tools designed for BIM-savvy users. This requires a process that assumes well-structured, complete models and relies on users to understand the model and rule logic. While helpful in theory, this approach doesn't align with how residential workflows actually operate. Here, users need support in detecting design conflicts, validating building codes, and producing stamped drawings for submission to municipal building regulators.
An essential aspect of this process has been to keep the engineers in the loop as part of the service, rather than letting architects use the AI directly as a co-pilot. Greenberg explains that this was important for ensuring quality and identifying opportunities for improvement:
We intentionally decided to keep Spacial-licensed engineers in the loop for three reasons: trust, accountability, and quality. In construction, it's not enough to be 'mostly right.' If a plan fails code or constructability checks, it's the builder and architect who pay the price. Cities don't want AI output. They require professional validation and plans to be stamped. By integrating licensed engineers into the loop, we give customers confidence that what we deliver is not just fast, it's reviewable, stampable, and real-world ready. It also allows us to improve our AI technology continuously, using real human corrections as feedback loops.
Part of this process involved ensuring that explainability remained top of mind across the types and uses of AI support. For example, every compliance check is linked to the specific rule it's derived from. The output plan also includes annotations and references to local code. Users and cities can see why a change was made or a component was flagged.
Most importantly, every set is reviewed by and signed off on by licensed engineers at Spacial. Greenberg says:
This creates a clear accountability chain: the AI proposes, the engineer verifies, and the customer gets a stamped, trusted result. It's not just transparency, it's responsibility, made scalable.
An essential aspect of this process has been to improve the ability to translate ambiguous legal text into machine-readable rules. Spacial has approached this with a hybrid strategy consisting of multiple specialized agents. Each agent is focused on a different engineering discipline and works with other agents and systems to handle reasoning, validation, and context alignment. Together, this helps translate complex city and building codes into a functional rules engine.
Spacial has also combined natural language processing (NLP) and classification models with an internal rules authoring team and QA process to guide this process. This includes teams working on jurisdiction-specific interpretations, written and reviewed by domain experts and code professionals. But it's not perfect. Greenberg acknowledges:
We don't try to turn every regulation into pure logic because much of the law simply isn't written that way. Instead, we triage: automating the rules we can express with high confidence and flagging the rest for human judgment or verification.
Thus, Spacial is gradually expanding into new markets, which has helped them refine this hybrid approach for the nuances of new jurisdictions. The AI agents continue to learn to interpret legal and engineering matters more precisely, while expert oversight ensures the outputs remain accurate, explainable, and compliant, even as local codes evolve.
The compliance management tools and services for architects, builders, and permitting agencies are currently evolving within different silos. The end result is that the logic, checks, and considerations in one part of the process are often lost to the next. Greenberg says:
Compliance is fragmented. Architects use one tool, engineers another, and permitting agencies often still rely on PDFs and emails. There's little visibility across stakeholders, which creates delays, miscommunication, and rework. We believe the opportunity is to create a shared compliance layer, one that integrates design, engineering, and code validation in a single workflow that's visible and explainable to all parties.
Spacial is starting with plan generation because that's the bottleneck. But in the long term, Greenberg believes there is an opportunity to build or collaborate on a platform layer that connects architects, engineers, permit reviewers, and city inspectors through a shared source of code-aligned truth, with audit trails, version control, and validation baked in.
Just like writers making different categories of grammatical errors, architects and designers can make a variety of types of mistakes. Common ones include floor plans that are not fully dimensioned and sketches that lack MEP layers. In addition to code and constructability, the process also helps analyze for cost-efficiency, timeline optimization, and overall design quality, so the output is practical and builder-ready.
To address these issues, the process starts with plan enrichment. This involves inferring and classifying building elements, then cross-checking against zoning data, code databases, manufacturer specs, and industry best practices to fill in the gaps.
A second step is to run a multi-pass validation on the enriched model to check the structural, mechanical, and regulatory layers for inconsistencies, missing data, or design risks. Anything that does not meet this automated quality check is flagged for human review before moving forward.
Building a better process for the construction industry faces headwinds from the industry's natural inertia and interoperability challenges, despite various standardization efforts. Greenberg argues that driving the adoption of more automated digital processes will require ensuring that no additional work is required and that no control is lost.
One aspect of Spacial's strategy has been to integrate directly into the tools architects already use, such as Revit and AutoCAD. They can upload their designs and get back stamped, permit-ready sets without changing the way they work.
On the regulatory side, they are partnering with permitting agencies and code consultants to align the output with city review expectations. This improves processes for submitting structured submittal packages with traceable logic and reviewer-friendly formats. Greenberg says their long-term strategy is to embed into the compliance lifecycle, not sit adjacent to it:
That's how we build trust, and how we become the infrastructure layer for modern permitting. The major gap is still semantic, translating vague or subjective code into clear, enforceable design constraints. That requires collaboration across tech vendors, local governments, and AEC professionals.
We need shared rule libraries, standard APIs for plan submittals, and digital audit trails that cities can trust. Spacial is helping lead that transformation by not just building AI-powered automation, but delivering it with accountability so builders, designers, and reviewers can all move faster without sacrificing trust.
One of the more interesting aspects of this is not just about improving construction, but how Spacial is exploring how to use AI to build a better process rather than make the old one faster.
I started investigating automated code compliance shortly after hearing about all those data centers NVIDIA planned to build in the UK for physical AI. It got me wondering how physical AI was being used to accelerate the construction of these planned data centers, in line with very optimistic timelines. I was surprised to see little progress in automating code compliance since about three decades ago, when I listened to builders complain about how hard it was to work with architects, and architects complain about the building regulators.
Data center construction is, of course, a whole different discipline than building houses, and you think maybe they are doing a better job at that. Some of the more visionary folks in the field certainly see an opportunity to streamline the legacy processes used to ensure code compliance.
In the long run, the real value of better processes will be realized when architects, builders and regulators all modernize their processes for the AI age. But this will require some balance and collaboration. For example, what happens if regulators make their building code more computable but also more brittle and prone to incentivize new problems that pass compliance checks?
I think that Spacial's approach of starting by improving the process, guided by experts in the loop, is a good first step toward bringing safe innovation to the architecture side. In the long run, it will provide a foundation for regulators in jurisdictions to also safely experiment with more computable building regulations. This could make it easier to sign off on new plans quickly. Down the road, it would also provide a foundation for streamlining building inspections by automating processes around capturing computable models of facilities as built with 3D capture tech. Spacial is currently focused only on residential construction in a few key markets in the US. It seems like there is room for more of this kind of process-focused innovation in other markets and for other kinds of construction as well.
A team of researchers from the UK, New Zealand, and Germany recently examined current progress in automating regulatory processes. They observed that while automation is certainly helping, there is also a growing shift towards performance-based processes in many countries, which adds complexity to automation. This might require developing more practical approaches, such as traffic-light systems and heatmaps, rather than binary past/fail results, to navigate these challenges.
Also, regulators in Singapore have proposed ways to capture the reasoning behind human decision-making. For example, the reasons waivers are granted should be documented, allowing ACC systems to learn from past decisions and propose similar waivers in the future. They also argue that innovations in ACC would guide improvements to building codes that simplify computability.
