i'm chris. i run anvol, an AI-native hardware contractor. you bring an idea, a design or a prototype, and we answer for the whole product, from engineers to factories.
@seveibar is the scan on the bare board or with cables plugged in? i'd want to know if the autorouted traces even show up next to what the cables radiate
AI is about to produce a lot more hardware designs. someone has to stand behind the finished device, not just their piece of it. that's what we sell at anvol: one contract for the whole product.
if an AI-designed board fails, who pays to fix it?
we read the fine print this week: an AI design tool's terms, an assembly shop's warranty, a typical design contract. all three point the same way, and it isn't at themselves.
@boardyai yes please, would love the intro. low-volume runs out of china are exactly what we do. my DMs are open, or the founder can reach me at [email protected]
i'm chris. i run anvol, an AI-native hardware contractor. you bring an idea, a design or a prototype, and we answer for the whole product, from engineers to factories.
@maahirpanchal the part i worry about is the 100-1,000 unit runs. small shops are the ones who'll actually quote a pilot batch, if they get absorbed i'm not sure who picks those up
i post about what breaks when you build hardware, and about running a company on agents. if you're a software team about to build your first device, my DMs are open.
@maxkolysh i run my inbox and lead finding on agents. the only part worth having is the work i'd never get to myself, like finding leads. anything that reports on mail i already read is just a second inbox
physical AI startups raised around $47B in the first half of this year. money isn't what's stopping hardware right now.
what stops it is the step after the prototype. getting the first few hundred units out. factories want big volume, the online platforms stop at metal and plastic parts, so the founder ends up stitching boards, firmware, enclosure, assembly and testing together by themselves.
that step is what we do at anvol. you bring a design or a prototype, we answer for the whole product with our engineers and factories. an agent will take that over piece by piece.
Most AI-for-hardware tools try to generate designs. I built Anvol to do the opposite: you tell it what you want to make, it tells you what to leave out, then engineers in Shenzhen build what's left.
The worst place to start a company is with an idea.
If it's genuinely new, no one is paying for anything like it yet. You'd need to invent the category, educate the buyer, and prove the pain — before building anything.
1000 recent YC and HN startups. 80% of the ones worth testing had one thing: someone already paying for a worse solution.
So I'm not starting with an idea. I'm starting by finding out what people in hardware already pay for — and where it hurts most.
Every AI creation tool is built on the same bet: make creating easy enough and the products will find their buyers.
We built one for physical products — one sentence in, a finished-looking hardware design out, a path to a real production run. We put it in front of about twenty friends. They had fun generating things. Not one of them actually wanted to take a product to market.
That's the era we're in: an idea plus the ability to manufacture it is worth nothing on its own anymore — AI made both cheap. Demand, traffic, someone already waiting to buy: that's the scarce part. So I'm going where the demand already lives — creators with an audience — and bringing them the factory.
If that's you — people keep asking to buy the things you make — my DMs are open.
i spent a month building things: firmware tooling, a way to catch claude
hallucinating in embedded code, a pile of skills, a modular pcb.
none of that was a company. i was building, not selling.
a startup sells something, or at least asks someone what they would buy. i had
done neither. so i stopped writing code and built a cold email system instead,
and sent about 200 emails to people who publish open hardware designs. three
questions: would you pay someone to manufacture this, do you want help building
it, and what is the part that ai still cannot get you past.
37 wrote back.
most of them had never built the thing they designed. a finished design,
published and documented, and zero units in the world.
the ones who did try and then stopped are the group i did not expect. not one
of them stopped because of the design. one stopped when a connector got too
expensive to keep sourcing. one cannot get the chips at all - the parts have
not been manufactured in years. one had it working and stopped shipping to the
us over tariffs. one is stuck on air-shipping a lithium cell sealed in a resin
case.
three of them, unprompted, offered to let me manufacture and sell their design.
in 37 replies, nobody asked for a better way to design hardware.
that was the entire premise of what i had spent the month building.
the tentative takeaway: the design is the part that already works. what nobody
has is the handoff.
i should have done this first.
we are widening the customer research now - more segments, more questions, and
no code until something comes back. i do not have the answer yet.