Autodesk Fusion + ai assistant

ReddZed · 2026-04-18T00:31:35+0100

Since I can barely draw a straight line, thought I'd better enlist the computer for some help. I'm a software engineer by trade so how hard can it be...

Sorted out a personal/non-commercial free license for Fusion without much fuss.

Within minutes of installing it, I'm reminded that I absolutely hate learning new modelling tools and how vertical the learning curve usually is :doh:

I'm modelling my welding table project (currently a 10mm steel plate on top of an existing workbench), and it was just too painful trying to work out how to make 4 angled steel legs to start with.

Then I noticed it had Autodesk Assistant, which appear to operate exactly the same as software engineering Ai assistants, which I have more than a passing familiarity with.

So I start telling it about the table I want: 10mm plate, 850x550mm, 4x 3mm angle steel legs

It asks for some clarifications on the position of the legs and the details of the angle steel, and suggested some common sizes

Then it built for me :clapping:

Then asked it to add a shelf at 300mm height, 2mm steel sheet with 20mm lip
We got into a bit of a heated debate as it kept drawing a solid 20mm plate, but eventually we got it right...not bad for half an hour's work in a new modelling tool!!

Then it asked if I wanted to add an apron rail to support the table top, or any bolt holes...went for the apron rails:

It made some mistakes, the rails were off to the side of the table initially, but it managed to spot that on it's own and recalculated the positions for them based on the existing model with no further prompts - I was quite impressed with that aspect of it.
It's not "instant" as the Ai is generating scripts to control Fusion, then executing those, then trying to verify results...but it was a damn sight quicker than I could have modelled it !!

Anyway, I just wanted to share this because it's a complete game changer for my modelling work...and if you've not tried it out, I strongly recommend doing so.
Sure, Ai is killing the planet and all sorts of other bad things, but so long as you give clear instructions and verify what it's done, it takes the drudgery out of CAD/3D modelling :clapping:

skotl · 2026-04-18T00:59:19+0100

Interesting. I saw the article on Ars Technica earlier in the week about Autodesk embedding agents into their products but for some reason Fusion didn't register with me.
I've been using Fusion for around 3 years and have got pretty (hobbyist) good with it but agree with you on the learning curve - and I still only know 5% of what it's capable of.

But I'll definitely give the AI agent a go - I'm pretty much relying on Codex now (ChatGPT) for software engineering and, as you've discovered with Fusion (and probably in your day job, too), as long as you erect the guard rails and review its progress, it is amazing as a productivity booster.

That said I'd definitely say that it makes an experienced person (whether that be software, hardware design, or whatever) more productive - it does not make an inexperienced person an expert.

ReddZed · 2026-04-18T01:18:49+0100

For software work, we're a Co-Pilot shop with a range of different models to choose from.

I've been a disciple of Test Driven Development (TDD) for over 20 years, and it's the only way to work effectively on software that's going to live for a long time and be updated frequently.
It turns out, if you use the test as the prompt, it's incredibly powerful and accurate...but yes, you do need a fairly good understanding of how to build software in the first place to craft those prompts.

Of course, the cool kids are all writing agents now...to control cohorts of other agents. Someone demo'd me their JIRA integration system where one agent scraped acceptance criteria from the ticket, while another one analyses the relevant codebase and starts suggesting changes, while another agent sets up a test environment so it can verify any changes. When it works it's very impressive, but needs tuning as it has a tendency to go a bit mad and solve completely the wrong problem.

skotl · 2026-04-18T01:43:03+0100

ReddZed said:
For software work, we're a Co-Pilot shop with a range of different models to choose from.

I've been a disciple of Test Driven Development (TDD) for over 20 years, and it's the only way to work effectively on software that's going to live for a long time and be updated frequently.
It turns out, if you use the test as the prompt, it's incredibly powerful and accurate...but yes, you do need a fairly good understanding of how to build software in the first place to craft those prompts.

Of course, the cool kids are all writing agents now...to control cohorts of other agents. Someone demo'd me their JIRA integration system where one agent scraped acceptance criteria from the ticket, while another one analyses the relevant codebase and starts suggesting changes, while another agent sets up a test environment so it can verify any changes. When it works it's very impressive, but needs tuning as it has a tendency to go a bit mad and solve completely the wrong problem.

And then we wake up in ~~the distant future~~ six months time and realise there are critical bugs in the system and nobody can decipher what the hell all the AI agents wrote.
I read a comment earlier today where someone was boasting that their AI agent had written 1 million lines of code in a month - the fact that some people can't realise that this is a bad thing is the biggest concern