Would You Drive Over An Ai Designed Bridge?
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━

No, not in 2026.

Date: October 28, 2026

No, not in 2026.

1. it does not exist
2. that scares the shit out of any sane person

I’ve had this talk with a couple of people that get it and seem to agree, but maybe this is just me ranting.

But
───

The time is coming.

The writing is on the wall.

In this round of model drops (oct 2026) Astra just pwnd every other lab with its insane modeling skills. These things are being RL’d on the mad right now to beat them at this benchmark.

I know, software engineers have seen this before.

Oh look at the cute little gpt4o making 20 line bash scripts for be because I dont wanna. I got better things to do than to learn bash and get things right.

Only a few years later we get opus4.5 and holy shit it set the world on fire. It could actually do real work without making twice as many bugs as features it added.

TESTING
───────

llms especially the openai models are pleasers, given a task they want it completed. Given a set of hard tests that it can go iterate on it will work to completion.

We have so much structural engineering science in this example, really the example applies to so much more, so much software that can already look at stress and fatige loads of structures. We have data for loads given the areas weather, usage, soil, all of the things. These things will learn how to iterate and run so many tests to make sure that structure is sound faster than any human can.

The human’s job is still important, if not more important. Many people probably got into that industry because they wanted to build cool stuff, and found a love for CAD, and they will be shocked, apauled, and go through all levels of grief when it does their CAD for them.

They will be able to step back and have the ability to ask so many more questions of their design than they thought were possible, running more simulation, and optimizations where it can be done by a fleet of agents without a human.

They still lack taste
─────────────────────

Anthropic models seem the best here, though kimi can do some dang good writing. As much as you want to believe they wont design the bridge before they can write a good proposal you got it backwards.

In the design phase here there are solid engineering priciples and decades of CAD and FEA software to drive. These are things we can goal seek to solve a problem. proposal writing, the look of the bridge, how someone might read text printed in documents or on physical labels, there is no test for that text like there is for the structure.