✉ Envelope #54: AI’s "Tools" - Potential Game-Changer for Engineers

Author

Andy Lin
May 30, 2025

Good morning! Andy from Back of the Envelope here.

If you've been following along, we've chatted about how Large Language Models (LLMs) and Large Reasoning Models (LRMs) work (link and link).

An LLM like ChatGPT-4o "predicts" answers based on patterns it's learned, while an LRM like ChatGPT o3 or o4-mini breaks down complex questions into smaller steps, carefully "reasoning" before giving you an answer.

As impressive as these models are, they've got some limitations:

  • Knowledge Gap: They only know what's in their training data, which stops at a certain point. (For example, if you ask, "Who’s the current president? Don't search the internet," it'll confidently say Joe Biden, because its training ended before 2025.)

  • Hallucinations: Sometimes, LLMs confidently give plausible-sounding (but totally wrong) information.

  • Math Struggles: They don't really "get" numbers and math operations the way calculators do, causing mistakes even with simple math.

But here's the good news: Researchers and companies like Google and OpenAI are actively working on these issues. A big part of their solution: giving AI new "tools" to enhance its capabilities.

Let’s dive in!
(Est. reading time: 2 min + 15 seconds)

What Exactly Are AI "Tools"?

When we say AI now has "tools," we're talking about its new ability to reach beyond its built-in knowledge and interact with external software, databases, and systems.

For instance:

  • Ask about current events, and it might actually "search the internet" for real-time answers (remember when that wasn't even a thing... like waaay back in 2024?).

  • Ask a math question, and it can spin up Python code to get you a precise answer.

Quick examples:

Without Tools (Non-Reasoning Model): Math errors often pop up.

With Tools (Python): Accurate, spot-on answers

Reasoning Models Automatically Use Tools: No explicit instructions required

Think of this like Excel functions… using something like FORECAST.LINEAR makes linear interpolation easy and accurate. Similarly, these AI models now tap into external "tools" for better reliability.

And it’s not just math, AI can now read PDFs or analyze uploaded images as well (something I am really excited to talk about — another time though).

Interacting with External Tools

Now here's the cool part: the trend these days are people building something called "AI agents". Think of these as specialized assistants that enable LLMs to directly communicate with software and tools we already use every day, making AI even more practical and powerful.

(It’s much more than that but this is part of it)

Imagine your AI:

  • Interacting directly with Google Calendar.

  • Fetching YouTube transcripts (and explain to you why that Minecraft video might not be appropriate for your 7 year old…).

  • Reading and writing ETABS data.

  • Pulling information directly from Revit models.

The details behind these integrations can get pretty technical, but the bottom line is: as companies provide AI with access to more external and internal "tools," its real-world capabilities will grow dramatically (even if the underlying models themselves don’t change much).

What Does This Mean for Structural Engineers?

As integrating these tools becomes smoother, I believe structural engineers can soon expect AI to seamlessly interact with tools we already use, clarifying assumptions, spotting missing info, and proactively checking our designs.

A few practical ideas that I thought of:

  • Quickly verifying load cases and combinations in RISA to make sure nothing slips through the cracks.

  • Cross-checking architectural plans against our structural models to ensure live load criteria are accurate.

  • Confirming boundary conditions in structural models match our design intent exactly (think fixed vs pinned).

These possibilities could be just around the corner give how fast the AI industry is advancing.

Alright and that's it for today!

What other ideas do you have about how AI "tools" could boost our work as structural engineers? I'd love to hear your thoughts!

Best,

PS.

If you are working on a steel building and you haven’t heard of Durafuse, you might want to check it out.

It is a proprietary steel moment frame connection that offers high ductility and fast recovery after an earthquake.

And they’ve got your back with full design team support related to the connections: calcs, plan check comments, RFI responses, and shop drawing reviews.

Click here to learn more: https://go.sehq.co/durafuse

Reply

or to participate.