✉ Envelope #55: I'm obsessed...

I'm obsessed... with coding using AI.

(Sorry for the clickbait-ish subject line.)

But seriously... coding with AI is getting ridiculous. It’s becoming incredibly easy and incredibly challenging at the same time.

On one hand, I’m barely writing any code myself. AI can do it 10x to 1,000x faster than I can — especially since I’m a beginner. Tools like OpenAI Codex or Google Jules can even run in the background and fix things while I’m off doing something else.

On the other hand... making a usable website or app is way more than just "coding." There’s version control. Databases. Tests. Security. Scalability. Usability. Architecture (different kind from the one we are used to). Deployment. And more.

All of these are things I’m just starting to wrap my head around.

And it kind of gives me a unique lens into what our profession might look like once non-engineers can "do structural engineering" with AI.

Let me explain.

Coding with AI vs. Engineering with AI

An amateur AI-enabled coder like me can now get a draft web app up and running. But...

• It’s going to have security flaws

• Trying to add features will probably break things

• The setup might not be right for the job

• And when things do break, I won’t know how to fix them quickly

Similarly, in the near future, an amateur using AI might be able to spit out a draft framing plan. But...

• They won’t understand how the details fit together

• They won’t know why some things work and others don’t

• They won’t catch subtle judgment calls

• Renovation? Good luck deciphering the as-built (if they even exist)

• And of course , they won’t (and shouldn’t) be stamping anything

Things will evolve as the tech advances...
But we’re still far away from AI replacing the full depth of what an engineer actually does.

It’s one thing to go from zero to a decent-looking prototype.
It’s another to make that prototype robust, secure, and usable.

Same in our world.

Getting from a blank sheet to a polished SD set might soon be doable by anyone with AI.
But turning that into an actual buildable CD set? That’ll still take experience and engineering judgement.

A Small App Experiment: ACI 318-14 vs. 318-19

With that said, I’ve been doing a lot of experiments lately.

Here’s one that actually worked, using Gemini (deep research) + Firebase Studio + Jules.

Remember how ACI 318-19 updated the shear equations and suddenly you needed a lot more reinforcing? (I talked about it here)

I built a little app to compare the cost difference with ACI 318-14.

I added two more options too:

• one where you boost flexural steel to get shear to work,

• and another where you increase member depth instead.

Then it compares all three using different f'c values to see which is most economical.

You can even click in and see the full line-by-line calc.

Check it out here and let me know what you think.

Took a lot of trial and error to get that experiment running and working.

I’m thinking of putting together a running log of all my experiments… wins and fails. Perhaps something will spark an idea for you too. Stay tuned.

Alrighty, that’s it for now.
Have a great weekend!

Andy Lin
Back of the Envelope/Structural Engineer HQ

P.S.

I’ve also been experimenting with using AI to generate memes based on the email I just wrote… but it’s kind of like asking the Terminator to write jokes. They’re not funny (at all), and it has to explain what the meme even means.

But somehow, with that context, the whole thing becomes slightly more humorous, in a dad-joke-ish kind of way.

Here is one (scroll down for explaination):

AI’s explanation:

My thought… yes of course, a hardhat and blueprints in the background will present the complexity of using AI in structural engineering!