What AI-Augmented Engineering Actually Looks Like Here

Everyone says they use AI now. It's become background noise — a line on a website that means nothing, because it's on every website.

So here's the concrete version. Where agents actually sit in how we design, build, and ship hardware. Three years in, and honest about it.

Start with what it isn't. There's no autonomous system designing products while we sleep. Hardware doesn't work that way, and anyone selling that is selling something. What we have is augmentation: a small team of generalists, each one covering ground that used to need a dedicated specialist, because an agent closes the gap.

Design

Schematic and layout review runs with an agent in the loop — a second set of eyes that never gets tired on page fourteen, cross-checking nets, footprints, power budgets, and datasheet parameters against the design. It doesn't replace the engineer's judgment. It catches the thing the engineer would catch on a good day and miss on a Friday.

Firmware

We reuse aggressively across products — the same modem manager, the same OTA pipeline, the same wake/sleep logic moving between an air-quality instrument and an industrial controller. Agents make that porting fast and traceable, and they read the sibling product's code before we redesign anything, so we stop re-solving problems we already solved.

Everything around the engineering

This is the part nobody talks about, and it's exactly where running a full ODM breaks small firms. Datasheets. Product briefs. BOM cross-checks. Test procedures. Customer-facing documentation. The website you're reading — the product pages, the structured data, this post. That's a marketing department, a technical writer, and a documentation team's worth of output, produced by the same people doing the engineering, because agents carry the load.

That's the whole argument from where I sit. Running as a full ODM means accounting, procurement, customer management, reviews, documentation, production — layers of function stacked on top of an engineering team. That was always a headcount problem, not a competence problem. Every small shop knew the fix and couldn't staff it. We're staffing it with agents. That's the shift I wrote about in the last post; this is what it looks like in practice.

What's next is the floor

We've pushed augmentation through design, firmware, cloud, and documentation. The frontier for us is manufacturing — a micro production cell per product line, with the same augmentation on the build and test side that we now have on design. Test-fixture generation, yield analysis, procurement, traceability. The parts of production that scale badly with headcount are precisely the parts agents are good at.

None of this is speculative here. It's how the two programs we shipped this year got shipped, how the datasheets got written, how this site got built. The ultra-specialist era is winding down — not because specialists got worse, but because a generalist with the right tools now covers what used to take a room.

The specialist didn't get worse. The generalist got tools.