Why AI Adoption Lags in Structural Engineering

In March 2026, Anthropic published a report comparing what AI is theoretically capable of doing in a given field against what it's actually being used for in practice.

Architecture and Engineering stand out in this report as having a large gap between theoretical use and actual use. Theoretical capability sits at around 85%, suggesting a majority of tasks could in principle be sped up by an LLM. But observed exposure was among the lowest of any field studied.

If this is accurate, engineers have one of the largest gaps between what AI could do and what it's actually doing. I think there are a few reasons for this.

Types of Tools

Many of the tools I see marketed are building code review or other types of code knowledge systems. The problem is, while these tools work just fine and are very efficient, code reasearch is just not an every day engineering task. Also, the nuance of the code is important, and while an AI summary is helpful, we ultimately need to read the raw code section ourselves.

Drawing readers and reviewers are another popular one. These are impressive tools that can significanly help with QA/QC of drawings. Most engineers should already be performing these tasks, but AI might be better as long as the drawing context is not too large. The engineer will still need to review the AI tool results. The value in this type of tool is more that it helps improve quality, rather than efficiency.

Then there are generative structure and form tools that create framing layouts or optimize a structural system. The output looks great in a demo. The problem is that real structural design isn't a clean optimization problem. The "best" layout depends on a lot of factors and coordination with other designers and contractors. These might be great preliminary tools, but they won't improve the efficiency of most projects.

3D modeling assistants run into a similar issue. It works great on a textbook homework problem type of building, but anyone who has created and managed structural analysis models for existing or even new structures knows there are a lot of engineering decisions on modelling that go into it. Handing off an entire model to AI is not very transparent and difficult to trust.

Aiming for Replacement

Many AI tools take the concept of replacing an engineer to heart as the goal. The problem is they target tasks like 3D modeling, framing layout, and full calculations from top to bottom of a project. Basically they want you to hand over drawings to the AI and get a completed project back.

The thing is, you wouldn't expect a junior or entry-level engineer to work so broadly and independently, so why start there with AI? A junior engineer is given well defined tasks and works their way up as they prove out their judgment. There's no reason AI shouldn't earn trust the same way, and we shouldn't expect AI to immediately step in at the senior engineer level.

Concern Over Safety

The broader the scope you give AI, the wider the range of solutions and the greater the risk of hallucination and mistakes. LLMs are improving their math and engineering skills, but they may not be reliable enough for life-safety work, at least not without a structure around them that makes review practical.

If an AI hands me a full set of calcs for a project, I now have to check everything in detail, every assumption, every load path. At some point that's more work than just doing it myself. The tools have to leave the engineer in a position where verification is actually faster than re-doing the work.

Clearspan

Clearspan is targeting a different approach:

• Start with simple, repeatable calculations rather than large projects.
• Let the AI work on smaller, well-defined tasks for better reliability.
• Gain efficiency by letting AI manage the repetitive tasks and calculations. Small wins collect into meaningful efficiency.
• Give the AI access to tested tools that it can use in a structured manner.
• Use tools engineers can open and run in place of AI. This creates familiarity with the surface the AI sees for better review.
• Create tools with well-documented output for thorough but efficient review.

The goal is a stack of small, reliable tools that an engineer can trust, verify quickly, and lean on for the parts of the day that were never the interesting part of the job to begin with. For an example of the kind of output these tools produce, and a verification example, see our wood beam design verification.