A Los Angeles startup says artificial intelligence handled most of the design work on a complex dual-PCB system, shrinking a three-month engineering task into just seven days.
A Linux computer with 843 components, spread across two printed circuit boards, was designed, assembled, and powered on in a single week. Even more surprising, it booted Debian on the first attempt.
The project wasn’t rushed by a massive engineering team. Instead, most of the work was done by AI.
A week-long build that usually takes months
The system, called Project Speedrun, was developed by Quilter, a Los Angeles-based startup focused on automating hardware design.
According to the company, the entire computer was designed and brought to life in just seven days. In a traditional setting, engineers say a project of similar scope would typically require around three months of focused human effort.
This wasn’t a stripped-down demo board either.
The machine includes 843 individual components laid out across dual PCBs, forming a complete Linux-capable computer rather than a simple prototype.
One sentence sums it up. This was real hardware, not a concept sketch.
Debian booted on the very first power-up
Perhaps the most eyebrow-raising detail is what happened when the system was switched on.
The computer booted Debian successfully on its first attempt.
Anyone who has worked with custom hardware knows how rare that is. Missed traces, timing errors, power issues, or simple layout mistakes usually mean multiple re-spins before a board behaves.
Here, that didn’t happen.
Quilter says only minimal human help was required during bring-up, suggesting the AI-driven design avoided many of the common pitfalls engineers expect.
It wasn’t perfect, but it worked. Right away.
Humans still mattered, just far less than usual
Despite the headlines, this wasn’t a case of engineers being removed entirely.
Quilter says humans spent a total of 38.5 hours supervising, reviewing, and refining tasks across the week-long build. That includes oversight, validation, and final checks.
The heavy lifting, however, was handled by AI.
Tasks such as iterative board layout, design corrections, and cleanup were automated. These are often the slowest and most mentally draining parts of hardware work.
One short line makes the difference clear. Humans guided. AI executed.
How Quilter trained its AI differently
Quilter says its system was not trained like typical large language models such as GPT-5 or Claude.
Instead of learning from human-designed circuit boards, which often include inefficiencies or mistakes, Quilter trained its AI using the physical laws that govern electronics.
That means voltage constraints, signal integrity rules, thermal limits, and other real-world behaviors shaped the learning process.
In simple terms, the AI didn’t copy people. It learned how circuits actually behave.
That approach helped it avoid repeating human habits that slow down design or introduce errors.
Why dual PCBs make this more impressive
Designing a single PCB is already complex. Adding a second board increases difficulty fast.
Dual-PCB systems introduce new challenges:
-
Connector alignment and signal timing between boards
-
Power distribution across separate layouts
-
Mechanical fit and tolerance issues
Handling all of that while packing hundreds of components into a tight design usually demands experienced engineers and multiple revisions.
That’s what makes Project Speedrun stand out.
The AI handled those interactions while keeping the system functional enough to boot Linux immediately.
One sentence says it plainly. This wasn’t beginner-level electronics.
What this means for hardware development
Project Speedrun hints at a shift in how future hardware might be built.
If AI can reliably manage early design, iteration, and cleanup, engineers could spend more time on system architecture, performance goals, and safety checks rather than manual layout work.
That doesn’t remove engineers from the loop. It changes where their time goes.
For startups, this could reduce costs and timelines. For larger firms, it could speed up experimentation and custom hardware development.
Still, this is one project. Scaling it across different products, industries, and safety-critical systems will take time.
Skepticism remains, and that’s healthy
Hardware engineers are understandably cautious.
Booting once doesn’t guarantee long-term reliability. Thermal behavior, manufacturing yield, and real-world stress testing still matter. AI-generated designs will need the same scrutiny as human ones.
Quilter acknowledges this, emphasizing that engineers still supervise and validate the output.
The difference is pace.
What used to crawl now moves quickly.
And in an industry where delays cost real money, that matters.
A glimpse of what’s coming next
Project Speedrun isn’t about replacing engineers. It’s about compressing time.
From three months to one week. From hundreds of manual hours to a few dozen. From repeated board revisions to a first-try boot.
That combination explains why this project has drawn attention well beyond the maker community.
