Hot take: within 18 months, agentic AI will write better embedded C than most junior firmware engineers. Not because AI is brilliant at firmware — but because most firmware is more repetitive than we'd like to admit.
Everyone's talking about AI writing software. Almost nobody's talking about what it means for hardware.
I've watched agentic AI tools generate schematic blocks, suggest component selections based on real-time supply chain availability, produce PCB layout constraints, and even flag potential EMC issues — all from a natural language spec. Not perfectly. Not yet. But the trajectory is unmistakable, and the pace of improvement is accelerating.
The hardware design cycle is about to compress
Think about what a typical hardware design cycle looks like today. An engineer spends weeks selecting components, cross-referencing datasheets, checking availability across distributors, evaluating thermal characteristics, and making trade-offs between performance, cost, and manufacturability. Then they capture a schematic, review it, hand it to a layout engineer, iterate on placement and routing, run DRC, and hope they haven't missed a decoupling capacitor or violated a return path.
Now imagine an agent that can do the first 80% of that work in hours. Not replacing the engineer — augmenting them. Handling the tedious cross-referencing so the human can focus on the genuinely creative design decisions. Running hundreds of layout variations to optimise for signal integrity. Flagging single-source components before they end up in a frozen design.
The uncomfortable truth for hardware startups
Here's the uncomfortable truth for hardware startups: if your competitive moat is "we have experienced analogue engineers," that moat is evaporating. Not because the engineers aren't valuable — they absolutely are — but because the barrier to entry for everyone else is dropping rapidly. The teams pulling ahead are the ones combining deep domain expertise with AI-assisted design flows.
Hardware risk is different from software risk
But boards need to understand the risk too. And this is where hardware diverges sharply from software. AI-generated code that's buggy crashes an application. AI-generated hardware that hasn't been properly validated catches fire, fails EMC testing, destroys a product certification timeline, or — in safety-critical applications — puts people at risk. You can't A/B test a PCB in production.
The governance challenge is real and it's novel. Boards need to ask: what's our validation process for AI-assisted designs? Who signs off? How do we maintain traceability between the AI's suggestions and our final design decisions? These aren't hypothetical questions — they're the ones that will determine whether AI-assisted hardware design accelerates your business or creates liability you haven't accounted for.
What this means for your firmware team
The firmware teams that thrive will be the ones that embrace AI as a tool while investing equally in the skills AI can't replicate: systems thinking, safety engineering, deep understanding of edge cases, and the ability to debug hardware-software interactions that no model has been trained on.
My advice to founders: embrace the tools aggressively, but invest equally in validation infrastructure. Simulation, pre-compliance testing, automated design rule checking, and — most importantly — experienced engineers who can spot what the AI missed. The startups that nail this balance will be untouchable.
The ones that use AI without guardrails will feature in cautionary case studies.