Two to four resubmittals. Hundreds of senior hours that no one pays for. This is the math behind every fab, every data center, and every pharma plant being designed in America right now, and the firms doing the work are absorbing it on their own P&L.
The work A/E firms can’t bill for
If you run a project architect’s calendar at HDR’s mission-critical practice, or you stamp drawings at CRB’s pharma group, or you respond to plan-check comments at Jensen Hughes, you already know the rhythm. The egress study comes back with comments. The H-occupancy control area exceeds an exempt amount. The wall rating in the cleanroom doesn’t match the new FM Global spec. And the senior team spends the next 80 hours redoing analysis on a fixed-fee contract that has already shipped.
The AIA Firm Survey puts design fees on these complex projects at 6-15% of construction value, with mission-critical, healthcare, and lab work clustering at the high end.² Inside the firms running this work, industry estimates place code-analysis labor at up to 25% of total architect hours,³ performed predominantly by senior PAs, PMs, and code specialists at loaded billing rates of $180-$250 per hour.⁴
On a single hyperscale data center project, that resolves to roughly 5,000 hours of code work, or ~$1.2M of compliance labor before a single submittal is filed.⁵ On a fab or GLP-1 plant the code stack layers NFPA 318, FM Global, cGMP, FDA validation, and proprietary owner standards on top of base IBC. Here, the same number rises to $4M per project. A firm running eight to fifteen such projects a year, ends up spending $10 to 40M per year on code analysis alone, before any rework is counted.
Rework lands on the firm, not the client
A/E contracts are typically fixed-fee. When LADBS, NYC DOB, or Phoenix PDD returns the third comment cycle on a complex project, the senior labor going into redrawing, code-analysis updates, and narrative responses comes off the firm’s margin. Plan reviews on this complexity tier average two to four resubmittal cycles,⁶ with each cycle consuming 40-120 hours of senior labor⁷ which translates to $15K-$80K of unbillable internal time per cycle, or $50K-$300K per project.
When code issues are caught later, the change orders run 0.5-2% of construction value per incident.⁸ A meaningful portion of that lands back on the design firm through E&O claims, indemnification clauses, and reputational cost on the next RFP.
This is the work A/E firms can’t bill for. It’s also the work that’s about to grow fastest in volume.
The most code-heavy buildings in American history are flooding the pipeline
The capex backdrop is real: $660B in hyperscaler AI infrastructure spend in 2026, $640B+ in semiconductor investment since the CHIPS Act, and $30B+ in GLP-1 manufacturing capacity announced by Lilly and Novo Nordisk alone.¹ These are the most code-dense buildings the US has ever built, and the firms doing the work. HOK, Gensler, HDR, Page, CRB, IPS, M+W, Jacobs, Stantec, plus a long tail of code consultants like Jensen Hughes, RJA, and Howe are about to feel every percent of margin pressure in the numbers above, multiplied by every project in the backlog.
The bottleneck isn’t capital. It’s senior-hour capacity at the firms responsible for getting the code analysis right.
Why the tools you’ve tried haven’t helped
The AEC software stack has matured in two directions, but the middle is empty.
Geometry-first tools like Solibri, Navisworks, BIM Track, Revizto can handle BIM data but can’t reason over a code section. Rule libraries are hand-encoded, hard to extend, and don’t generalize across jurisdictions or owner standards.
Document tools, emerging submittal-review and specification-AI platforms reason over text but can’t integrate BIM or interpret it without hallucinating.
Code consultants themselves are the most over-subscribed resource on every project. They hold the judgment that gets you through a stamp, but they’re bound by their own hours. They can’t be in twenty jurisdictions at once for a hyperscaler’s reference design rollout, and they can’t review every reference-design revision in parallel.
What’s been missing is an agent that understands BIM model and the rule corpus together, and produces findings that hold up under stamp.
Introducing AEC Evals
AEC Eval is an AI agent that ingests a project’s BIM model and any compliance corpus you put in front of it.
Workflow
Upload or link your IFC. Select the codes you want to check against. Hit Evaluate.
Findings stream in as the agent works through the rules, each one pinned to the exact element in the model. Every finding carries its full provenance: the rule excerpt it came from, the measurement that triggered it, the value, and the threshold. Click a finding and the 3D viewer frames whatever the rule was reasoning about, alongside the underlying BIM properties. Evaluations export as JSON or as a formatted PDF report, with BCF export on the near-term roadmap so findings flow directly into your existing review workflow.
Reviewers stay in the loop throughout. Approve a finding, dismiss it with a reason, or hand it to the code consultant of record. AEC Evals runs the code reasoning including measuring, cross-referencing, interpreting against the rule corpus, so your senior team can spend their time on the judgment calls that genuinely need a stamp.
The current release ships with the IBC 2021 corpus and supports IFC2x3 and IFC4. More building code libraries including NFPA 318 and NFPA 75 are next on the 2026 roadmap.
Powered by InterpretAI, because compliance agents have to be right
AEC Evals is powered by InterpretAI, a team specialized in agentic Root Cause Analysis: a way to explain why an agent reached the conclusion it did, surface the failure modes before they reach a human, and give engineers and code reviewers a way to verify the output before it becomes a stamped drawing.
Every finding the agent produces is traceable end-to-end: from the rule excerpt it was reading, to the measurement it computed, to the model entity it ran against, to the version of the code corpus in use.
We’re not replacing the AEC stack. We’re the missing reasoning layer.
Why now
The construction industry has heard “AI is coming” for a decade. What changed in the last six months is that AI agents can finally read structured engineering data and reason over it with traceable evidence at every step. AEC Eval can now read a 50-page code chapter, interpret it, and run those rules against a multi-discipline IFC model with full provenance for every finding. That wasn’t possible in production until now.
That capability arrived at the exact moment the US economy started building the most compliance-heavy projects in its history.
We’re going to spend 2026 working with the A/E firms, GCs, code consultants, and SaaS platforms that are already doing the hard work. If you’re one of them, request early access below.