A senior embedded engineer at an automotive supplier received a mandate from their VP: starting Q2, the dedicated QA team is being eliminated. Every feature team will now own their own automated testing using AI tools, with "highest tier subscriptions to whatever we need" and no manual testing before release.

This post struck a nerve in the embedded community because it sits at the intersection of several tensions that are reshaping the industry right now:

• AI hype meeting safety-critical reality. HMI software running on QNX and Linux in automotive isn't a web app where you can push a hotfix in ten minutes. These are systems where a frozen screen or a misrendered warning indicator has real consequences for drivers. The gap between what AI testing tools can do today and what rigorous automotive QA demands is significant.
• The "shift left" philosophy taken to its logical extreme. There's a well-established argument that developers should own quality — writing unit tests, integration tests, and thinking about edge cases from day one. But there's a difference between developers contributing to quality and developers replacing a dedicated QA discipline entirely. Testing embedded HMI involves hardware-in-the-loop setups, visual regression across display variants, and interaction testing that requires specialized skills and infrastructure.
• Budget pressure disguised as innovation. The post hints at what many commenters likely recognized: this reads less like a thoughtful engineering strategy and more like a cost-cutting measure wrapped in AI buzzwords. Giving teams "highest tier subscriptions" to AI tools is cheaper than employing experienced QA engineers.

What makes this post genuinely educational is the window it opens into how organizational decisions ripple through embedded development. When a VP sends that email, the senior engineer on the ground has to figure out what it actually means: Do we now need to build and maintain our own test harnesses? Who validates the AI-generated test cases? What happens to our certification process? How do we handle regression testing across multiple hardware revisions?

The 22 comments likely contain a mix of commiseration and practical advice — exactly the kind of hard-won operational knowledge that doesn't appear in textbooks. For anyone working in embedded systems, especially in automotive or other regulated domains, this is a case study in how not to manage the transition to AI-assisted development workflows.