AI in Physical Product Development

Where AI helps in hardware product development and where it doesn't: ideation, market research, design, firmware and the rising cost of change.

By Nathan Ruttley, CEO and Co-Founder·8 April 2026·4 min read

AI has transformed the software product management process, but hardware is a different story.

I found myself nodding along to an article from Cat Wu, agreeing with the observations made about software development - how the boundaries between requirements gathering, design, and engineering have been blurred - but then I started wondering about how this applies in my world: hardware.

The key difference comes down to cost of change. The cost of change for software has fallen through the floor but hardware hasn't seen this same benefit - it compounds throughout the course of a project. Issues caught at the specification stage are usually simple (and cheap) to fix, issues caught at tooling or manufacturing scaleup can have serious financial consequences. "Ship fast and iterate" doesn't always work in the hardware realm.

So the question isn't whether AI has a place in hardware product development. It's where it fits across the development timeline - and the answer, broadly, is: the earlier the better.

Ideation and problem definition

We have noticed an increase in non-technical, solo-founders enquiring with us and a common theme that emerges is that they have used conversational AI extensively in briefing us about their project. This has its benefits (empowering anyone to engage with a product design agency), but also its downsides (LLM overconfidence and solutionising too early). For us this means we have to work a bit harder to demonstrate our value during the early stages of product development - thankfully it usually becomes clear why you need a few humans-in-the-loop when it comes to hardware.

These conversational agents have genuinely changed how early-stage product definition works, especially for solo-founders:

• You can now talk through your ideas in private and surface alternative viewpoints and hidden requirements.
• Nailing down the problem definition.
• No need to engage an external team or friends and family for that first ideation session (although there is still enormous value in this!)
• For non-technical people you have a thought partner who can ask good questions, but also stay on your technical level.
• User journey mapping can be iterated extremely quickly.

The biggest risk here is bad prompting. A bad prompt can cause the LLM to lead you down a dead-end path, over encourage you to pursue a "bad" idea, and start solutionising. And all of this will be done with the typical overconfidence of a conversational AI - cementing that idea and solution in your head without challenging you at all.

It follows then, that good prompting can give good results in the ideation phases of a project. I have created a Purple Pages doc with some prompt suggestions, you can grab it here.

Market research and competitive analysis

Still in the early stages, and still a relatively low cost-of-change zone, there are some other great use cases. You can utilise research tools to understand your addressable market size and explore secondary revenue streams. Analysing the competitive landscape that your product sits within can take a lot of time, AI can do that initial research for you by pulling together your direct and indirect competitors, gathering their pricing information, financial details, and marketing strategy. The key here is to ask for citations - and to check them!

Design: where humans still lead

As a project moves into design, the cost of change starts to climb - and this is where the limits of AI become most visible.

When designing for humans there will always be a human element to the design process itself. We have just described using a machine as a partner in that design process, not as a replacement for humans. When using AI to enhance the product design process, it leaves more space for humans to add value where they do best - feelings. Design is inherently about feelings: how it feels to have your problem solved, how it makes you feel interacting with a product, even what emotions you experience when looking at a product! Feelings are valuable throughout the design process too.

The value of a design agency is in the lived and embodied expertise from years of working with materials, manufacturing methods, tools, and of interacting with hundreds of people from all walks of life across many projects. LLMs don't have hands (yet) - they have never felt the difference between different surface textures, or understood how a torque setting actually feels, and they've never felt cosy or safe or excited. An LLM's understanding of physical reality is limited to equations and the descriptions of reality by humans. A bigger context window can't fix this.

Further down the timeline: firmware, components, and what's coming

Further into a project, where the cost of change is highest, AI tooling is least mature - but it is creeping in. We can already see benefits in the firmware space and component selection. There is a lot of hype around generative circuit generation and PCB layout but those products are nowhere near mature enough to use in a production environment... Yet. Personally I would like to see more accessible simulation tools, DFM analysis, and maybe a sprinkling of generative tools for some design grunt work.

Where does that leave us?

Map AI value against the hardware development timeline and a clear pattern emerges: the earlier the stage, the lower the cost of change, and the more confidently you can deploy these tools. Used well, AI currently adds huge value at the front end of a project. Using it well is an art and we shouldn't expect great outputs from poor inputs. But the real opportunity is in freeing up human expertise for the stages where it matters most - and where no amount of context window will replace a pair of hands that have actually built something.