Who will be left to solve the hard problems?

At every company there are generally always only 1 or 2 engineers who solve the tough problems when no one else can. There is no secret as to why…the bell curve of talent exists in engineering just as much as is does anywhere else. However, I would postulate that true troubleshooting ability is even rarer which makes “the guy” who solves every problem even harder to find. I mention that because that because true troubleshooting ability doesn’t seem to necessarily track with IQ or education. I have seen some very high IQ, book smart people who struggle to troubleshoot something as simple as the breaker being tripped.

It also generally requires a cross-discipline type of knowledge. Have you ever changed your oil, your tires, or even pulled an engine out of car? Hands-on, real-world knowledge adds greatly to the cause of complex troubleshooting. Unfortunately, less repairable products and fewer opportunities for kids and teens to get jobs limit this experience.

The number of people with this broad expertise seems to be dwindling and AI is only going to make this problem worse. The best example I can give currently is the push to go from bare metal firmware to everything being abstracted from the hardware in the form of libraries or worse an RTOS. I don’t want to be the “get off my lawn” guy because I recognize that with the cost/performance of modern MCUs, it is possible to quickly write many applications without really understanding the hardware at all. With Zephyr now emerging this is only going to get worse.

The problem this brings about though is that so many “FW engineers” are really “programmers”. They know the language and the various tools, but when a library function fails to work, they have no idea how to trace out the root problem.

As a recent example, a meeting produced the statement that “this I2C peripheral can’t be on the bus with other peripherals because once it is read from, the other peripherals don’t work”. I listened to variations of this comment for almost 9 months where the consensus was that an expert from the division that wrote the library call to this part will have to investigate. Finally, one day after not being able to stand hearing this conversation for what felt like the 400th time, I spent 2-3 hours writing a bare metal driver for the part. I already had written bare metal drivers for other parts on that same bus. Lo and behold, I’m able to communicate with all the parts on the bus including this “problem part” in any order without issue. Nine months of discussions around something that is not actually a problem. Now, I am not some high IQ savant…I’m simply a guy who spent a career continuing to learn and perfect my craft. I have worked with many super high IQ guys who can humiliate me with regularity…spouting off graduate level math in their head on topics I have long since forgot. But that does not automatically translate to troubleshooting and problem solving. It is a skill in and of itself.

Like any skill, it needs to be honed. The more we divorce engineers from understanding how things work all the way to the most basic level, the fewer versions of “that guy” will exist to come to the rescue.

AI and the loss of expertise

Love to hear feedback or questions