Here's Our First Gemini Deep Think LLM-Assisted Hardware Design

We've been using LLMs for software and firmware for years... now we're trying hardware. Threw a MAX44009 datasheet at Gemini Deep Think, asked for an EagleCAD library file, and about 10 minutes later it popped out working XML. Loaded it in Eagle, checked the pins and dimensions, rolled with it.

Correct pads, correct pin naming, even added a pin 1 dot and sensing element outline on tDocu. Ladyada's still driving... just has a co-pilot now 🎳

https://blog.adafruit.com/2026/02/14/heres-our-first-gemini-deep-think-llm-assisted-hardware-design/

@adafruit I am beginning to realize that I need to start a blog. This is giving me a lot of ideas related to my plans to create FOSH (Free Open Source Hardware) that depends only on FOSS (Free Open Source Software) and doesn't require any paywalled or proprietary tools for its design. However, all of my ideas recently come down to: I need to create a working FOSH product that people actually buy.
@neptune22222 @adafruit Have a peek at Oxide Computer, if you haven't already. For example,
https://rfd.shared.oxide.computer/rfd/0241
241 - Holistic Boot / RFD / Oxide

@davecb @adafruit

That's awesome to think about buying our own cloud computer to support our nonprofit. I really hope editing a PCB doesn't require a cloud computer. The PCBs that I'm designing do not yet require a cloud computer to edit the design...

@neptune22222 @adafruit

Nor do I!

What I found interesting is the degree to which they support open hardware and software.

@adafruit I like the use of the DFM tools in the middle. Would be interesting to study the timings of the pipelines to see if some of the steps could be optimised by asking the llm to write code for some of them?

Datasheet + LLM => intermediate format
then effectively compile that to the library format.

Good work, particularly using your own extensive collection of work as context for the AI

@adafruit This is sad. I used to have such respect for adafruit. I have lost it now upon learning this. #NoAi

@malachai @adafruit

Same here. I thought adafruit could be interesting for #Teckids tinkering projects, but now that's off the table.

@nik @adafruit They truly used to be. Lady Ada was one of my heroes when learning on arduino back in the olden times, lol

I'm extremely disheartened by this. I understand dabbling a bit in genai, but now that we know the societal and environmental impact... I can't support anyone - especially my heroes - who uses it seriously, especially for products they intend to sell, ESPECIALLY for simple products.

@malachai @adafruit I mean no disrespect when I ask why this is not an appropriate use generative AI? Capturing footprints and schematic symbols is pure drive work that can take hours before you even start the actual engineering. The tech bros talk a lot of crap, but getting the machines to do these boring robotic tasks really does free up engineers to do creative work.
@orangutanlibrarian @malachai @adafruit not really, managing the library and footprints is about 1/10 of thr work load of electrical design. Hardly ever you need to make your own footprint since most components use standard footprints. Realistically most vendors already offer kicad library files if not the footprint files. I would wager it takes me more work checking if the shitbox machine generated correct files then I am making my own files.
@tretron @orangutanlibrarian @adafruit yeah the slop machine requires a lot more work to stupervise than it would take to design that dinky little board. That's part of why I'm so disappointed.
@tretron @malachai @adafruit All of that is true, but at 10% time saving on projects with weird components. If, big if, the machines can be relied upon to produce footprints and symbols faster and more reliably than I can do it myself, then why not? Should we never use the tool just because it's mostly used for copyright violation and porn?
@orangutanlibrarian @tretron @adafruit that's not my issue with genai. My issue is that its burning the atmosphere and destroying the hardware market. #noai #noaislop
@orangutanlibrarian @malachai @adafruit To expand on Malachai's reaction too: There is also an issue with who owns the tool itself. Making footprints is a skill, reading documents is a skill. Those skills are important in electronical engineering to understand what you need to do. Now generative AI is a tool that generally is not owned by you the person but rather a large cooporation. using this tool would reduce the need for anyone to hone their skills. This causes a feedback loop where someone becomes dependand on the tool. Basicly a circle saw is a tool i often use, but I don't need a circle saw to make ca cabinet. I can transfer most skills over to my hand saws. It will take me a while to get a feel for the new tool and it isn't as fast but the underlying skill set is still the same. With generative aI as a tool that's not the same. It is basicly a new form of feudalism. That is ontop of the negative envionmental effects both dirrectly and long term.
@tretron @orangutanlibrarian @malachai @adafruit and let's not forget that this technology currently burned through 840 billion $ of aggregated funding. With a microscopic fraction of that budget we could probably automate the issue away with a suitable and reliable tool. For that kind of money every problem you can think of has a better solution than an LLM.

@gabrielesvelto (removed other respondants not to spam with messages)

I did a bit of estimation and 840 bilion $ would roughly equate to 56700 metric tons of CO2 reduced. Which would save human kind a collective 10,489,500,000 $ in climate related damages annualy. And ofcourse this does not take into account what such a boom of money would do to the reduction in price for green energy. so the effects would be even greater.

@tretron @gabrielesvelto one ton of CO2 removal costs less than €100 atm, so you're a few orders of magnitude off.
@iwein @gabrielesvelto I did not calculate with removal cost but the cost that the damages would do per ton CO2. as a guestimation, worked with about 195 USD per ton, would be roughly the same magnitude as 100 euros per ton. It's all guestimation in the end. Not a client scientist, they do the fancy math. It's more the point that all this money that is beeing poured inot Generative AI datacenters could honestly be better used reducing the harmful gasses.
@adafruit well that's too bad. Guess I'm not buying adafruit anymore. #NoAI
@cocaine_owlbear Unrelatedly, I love your handle lol
@malachai thank you 😊 it's nice to be appreciated
@cocaine_owlbear @adafruit Why can't these fucking companies JUST FUCKING STOP! 😡

@cocaine_owlbear Can't reply to them directly 'cause I've defederated fosstodon already...but I did send them an email, after which the account I've had since 2018 got deleted. ZERO tolerance for this bullshit anymore.

I'm a software dev. I've seen the kind of code these things produce. It'll pass your test cases and then fail catastrophically the moment you try to actually use it. And then they give it to me to fix it and it's like something out of an obfuscated code contest, damn near unintelligible, and the only solution is for me to do a full rewrite. If they're building their hardware like that and then just "rolling with it" then I can't trust a single damn product they produce...

And this is PARTICULARLY egregious for a company that mostly sells to amateurs who might not have the skills or tools to determine if it's a faulty component; they'll be sitting there thinking THEY did something wrong!

@admin @cocaine_owlbear wow. They retaliated for a valid concern? That's horrible.

@malachai @cocaine_owlbear No, *I* defederated fosstodon ages ago.

Edit: Because they were run by bigots. Zero tolerance for that shit too.
https://fediversereport.com/fediverse-report-114/

Fediverse Report – #114

Instance drama, with some reflection on how federation could be improved on the fediverse.

fediversereport.com
@adafruit i'm not buying from you if you're using ai slop to design your products
@adafruit good for you, and for me. I can now buy the Chinese clones of your hardware without a bad conscience.
@mavu @adafruit Why though? This is a perfect use for generative AI. Creating footprints is pure slog work, pad size, pad spacing, solder paste, solder mask, silk screen. Look up the the sizes in the data sheet, or a seperate sizing spec sheet if it's from TI. The associate pins with pads. It can take hours and at the end you have a schematic component so now you can begin the actual engineering.
@orangutanlibrarian @adafruit because genAI companies are anti-social parasites who only thrive by stealing everyone's creations without asking for permission or giving even a thought to compensation. From music to art, from books to shitposts. Everything stolen and the result sold for money.
They destroy lives of people from artists to engineers.
They eat energy and capital at an insane rate, and the crash will be a desaster.
Use genAI and accept all of the above as inevitable, or fight it.
@mavu @orangutanlibrarian @adafruit That's our capitalist economy. None of this is new or specific to genAI companies. The stuff the exact same system does in african countries is even worse. It's important to recognize this to better navigate how to make the world better. Exploitation, disregard of ownership and rights, erosion of freedoms, propping up corrupt governments, leveraging the power advantage the system provides for ill; all that you listed are subset to this, and are a property of a deeper older misanthropic engine that in an ideal world would not exist.
@adafruit ow no, don't jump the shark.
@adafruit Please learn to read the room instead of leaning harder into a space your customer base hates.
@adafruit
Oh, this ain't it. Neeeext.
@adafruit So what you're saying is you had a software tool assist you in stealing from designs made by other people, which you're now selling for a profit. Got it.

@scy @adafruit

I remember when people did not brag about stealing. Then the "only fools pay taxes" guy was elected President.

@adafruit Onto my blacklist you go ...
@adafruit so, the big deal is not the thing it does, but the fact that it's slop generated? That's what you're proud of?
@adafruit really, no. Expected better.

@adafruit but… why? You talk about how this piece was made but don’t actually describe what it does. Why would I want to buy this hardware? The only “advantage” you’re giving me is it was designed by AI. Which makes me not trust it, because AI is very imprecise.

AI sucks, you should feel bad for using such environmentally regressive tech, and you should feel embarrassed that your embrace of this imprecise, unproven technology shows your institution lacks critical thinking.

@sidereal @Em0nM4stodon @adafruit man I recommended adafruit on my website, guess its time to delete that part =(

@sidereal @Em0nM4stodon @adafruit @ArcMother

Don't just delete. Loudly un-recommend them and explain why.

@adafruit

But why?

@float13 @adafruit To let a machine do a tedious and boring task, translating a datasheet into a description file for the CAD software.
The creative stuff comes later, by humans.

@adafruit talking about genAI on mastodon is a risky proposition for a business (or anyone) heh. fwiw, parking the ethical issues of how commercial LLMs are made and pushed for a second, this might actually be a legit non-genAI use of LLMs, if they can do it consistently.

It does feel like cracking a walnut with a sledgehammer though - factoring in ethical considerations (wearing the rep hit, or trying to create a clean specialised LLM) would it be just easier to draw them the old fashioned way, or use a commercial provider for the parts model?

edit: I wonder if above is more technically accurate to talk about NNs rather than LLMs.