Summary: Most factory-automation startups want to build robot-dense facilities from scratch. Harmoni, which just raised a $10 million Series A led by Bessemer Venture Partners, is betting the opposite: that AI wins in manufacturing by upgrading the roughly 280,000 existing US plants that still run on legacy machines and human labor. For product builders, it’s a useful case study in why context, not model quality, is often the real deployment problem.
The wedge: a magnet-mounted tablet
Harmoni was founded in 2023, but the idea came from co-founder David Caputo’s first factory visit after leaving investment banking — workers lining up to clock in, then carrying USB drives loaded with machine programs between stations. The information on the floor was fragmented across systems that didn’t talk to each other.
Existing tools like MachineMetrics and Datanomix each covered one slice. Caputo and co-founder Adam Ellis, a longtime manufacturing consultant, started building in 2020. The core of the system is a tablet magnetically attached to each machine, connecting everything from brand-new equipment to WWII-era machines. Workers clock in with a badge, read work instructions, track job orders, and report issues on the same device.
The results at customer WessDel, an aerospace and defense precision parts shop in San Jose, are concrete: VP Jeff McKay says workers previously spent up to nearly an hour a day queuing at three machines and manually entering data; badge check-in and automatic machine logging eliminated most of that.
HAL: AI that already knows the context
Once the data was centralized, Harmoni shipped HAL, an AI layer that answers questions and takes actions directly on the factory floor. Its pitch against generic AI assistants is structural: a general assistant needs the user to explain which machine, which part, which past problems. HAL’s tablets have been sitting on the machines, so it already knows what’s running, what job is active, and what happened last week.
HAL does three things: diagnose (“why is this machine slower than expected?”), execute workflows (submit a maintenance request without paperwork), and hand off (flag issues for the next shift). The sequencing matters — Harmoni built the tablet first, the AI second, precisely because AI can’t understand a floor whose data is scattered across disconnected systems.
That’s a pattern worth internalizing. Whether it’s a shop floor or a log pipeline, deployment value comes from meeting the system where the context already lives. We’ve made a similar argument about robots that communicate through their environment rather than despite it — the interface design is the product.
Why the retrofit bet lands
Two tailwinds make Harmoni’s positioning credible. First, according to TechOrange’s report, Cal Poly Pomona’s Shokoufeh Mirzaei notes many small and mid-size factories lack even basic data analysis and run pre-internet systems — they won’t replace expensive equipment but will retrofit it. Second, labor: Deloitte and The Manufacturing Institute estimate US manufacturing may need 3.8 million more workers by 2033, with roughly half of those roles going unfilled.
That shortage shapes the product’s politics. Automation draws union concerns about job losses, but Harmoni positions HAL as cutting repetitive administrative work, not headcount — and notes many customers can’t hire enough people anyway. A tool that saves workers time in a labor-scarce market gets adopted; a tool framed as replacement doesn’t.
The takeaway for builders
The obvious comparison is greenfield players like Hadrian and Machina Labs building heavily automated plants. Harmoni’s counterargument, in Caputo’s words, is that its system exists to let existing factories compete with fully automated ones. On humanoid robots, he’s blunt: they could eventually be coordinated by Harmoni’s system, but the technology isn’t there yet — until then, factories depend on people.
The transferable lesson: when everyone in a market chases replacement, the company that makes the existing asset smarter often faces less resistance and a bigger installed base. Ask where your users’ context is already stuck before deciding whether to build something new or wrap what’s there.
One limitation worth flagging: Harmoni has around 40 customers concentrated in aerospace and defense, so the model is proven in a high-value niche, not yet across the broader manufacturing base.