In a recent episode of Electronic Specifier Insights, host Paige Hookway sat down with Valentina Ratner, Founder and CEO of AllSpice, to unpack what it really takes for hardware engineering teams to adopt AI effectively.
The conversation focused on what works in practice, where AI genuinely adds value, and what must be in place before any tools are even considered.
AI is powerful – but not a silver bullet
One of the biggest misconceptions Ratner sees is the assumption that AI should be applied everywhere, to every problem.
“AI has gotten so much attention, and it’s a very powerful technology, but not every use case is a problem for AI … it’s not a silver bullet, it’s not a fix it all.”
She stresses that some problems are still far better solved with traditional deterministic software or automation. The first step, she argues, is always to deeply understand the problem and then pick the right tool for the job – sometimes that will be AI, sometimes it won’t.
The pressure on hardware teams
While AI is often discussed in terms of models and software, Ratner reminds us that none of that runs without hardware infrastructure. The same teams designing boards, systems, and devices are now facing unprecedented pressure:
- More programs to deliver
- More complex boards and systems
- Much shorter timelines
She notes that many of the companies AllSpice works with are essentially powering the AI boom but don’t yet benefit from AI in their own workflows: “It’s crazy to me that these hardware teams are building the future of AI, and a lot of them don’t have AI for themselves … they are powering for others, but when it comes to their own workflows, there’s still a lot of room to grow and improve.”
The prerequisite: a strong data foundation
Before AI can meaningfully help, teams must fix their data foundation. AI and agents are only as effective as the data they can access.
“AI and agents are only as good as the underlying data that they have access to. And if your data is incomplete, unorganised, inaccessible … there’s only so much that the agents can do.”
Most hardware stacks were built for humans clicking through desktop tools, not for agents operating programmatically. Ratner highlights key groundwork:
- Implement good revision control
- Standardise data formats
- Centralise schematics, PCBs, data sheets, supplier information
- Pull scattered context out of email, Slack/Teams, and “analogue” artifacts like whiteboards and sticky notes
The goal is to create an environment where agents can actually read, interpret, and act on the information.
Capturing design intent: from analogue to self‑documenting
A major challenge is surfacing the reasoning behind engineering decisions – the design intent that usually lives only in people’s heads.
Ratner describes how high-performing teams move from sporadic, heavy documentation to self-documenting processes: “The most successful teams … implement what becomes a self‑documented process … every time you make these changes, it kind of gets recorded almost automatically for you in the background.”
Rather than one massive review and documentation push at the end of a design cycle – when no one remembers why a decision was made – teams break work into smaller, frequent reviews. This both improves quality and naturally captures intent along the way.
Getting the balance right: deterministic systems, AI, and humans
Ratner suggests thinking in three buckets:
- Deterministic systems – known problems with known solutions and clear pass/fail criteria
- AI – interpreting ambiguity, massive context, and repetitive, high-volume tasks
- Humans – known problems with unknown solutions, creative design, trade-offs, architecture, and intent
“No deterministic system and no AI is going to invent things … humans are great for the known problems with unknown solutions … that’s what makes engineers engineers.”
She warns that some teams push AI too far, too fast, into the wrong use cases and end up with negative ROI, while others hold AI to unrealistic, computer-like standards of perfection: “I usually say I hold AI to more human standards than computer standards.”
How to start: one use case, one team
For teams wanting to get started, Ratner recommends a phased approach:
- One use case, one team
- One use case, many teams
- Many use cases, many teams
The first use case should be inspectable and verifiable so engineers can maintain oversight. For most AllSpice customers, that starting point is design reviews, where agents can do 80-95% of the heavy lifting and humans review and approve.
The payoff, when done right, is significant: improved engineering efficiency, shorter timelines, fewer respins, and reduced cost and risk.
The future: never paying for the same lesson twice
Looking ahead a few years, Ratner imagines hardware teams where institutional knowledge is fully captured and reused: “Never having to pay for the same learnings twice … turning all of those things into living institutional knowledge and memory.”
In that future, engineers spend far less time on repetitive, tedious work and far more on architecture, innovation, and creating the next generation of hardware – supported, not replaced, by a team of AI agents.
To hear more from Valentina Ratner, you can listen to Electronic Specifier’s interview on Spotify or Apple podcasts.
