The pressure has never been higher to ship fast. The board is devouring every news article about how AI has 10X the productivity of everyone and their uncle. They are asking you "how do we do the same thing? How do we 20X it?" You know they aren't thinking deeply about it and getting caught up in the hype. You will come up with a solution or a reason why the hype isn't real, you always do. But either way, the adaptation to the shiny new toy is going to rest on your shoulders.
The good news is that you've been through these hype cycles before. DevOps, microservices, and no-code platforms all were supposed to be game changers. Leaders and investors got caught up in those too. They took the headlines and glamorized a future where their engineering leaders bolted on the new technology and realized outsized profits. You know that each of those cycles told the same story. The early adopters reported outsized gains, but always neglected to share the true costs. The market jumped on board with varying degrees of success, and overall there was a one-time bump up in efficiency on average. Of course, it brought on a whole host of problems that you and your peers had to solve.
This time is no different. However, the momentum behind it is different. It feels like everyone paying attention is surging with power. Everyone has a little more pep in their step and is seeing a bright future. The optimism is really wonderful, but you know that for every advantage, there comes a cost. When you look around, most of the headlines are screaming upside potential. You've had some conversations about the risks. Probably, not all of them, but you know that non-determinism of the system outputs is positively correlated with increased cost in validation and security. The board thinks the problem can be solved by throwing money at the risks as fast as the code is being generated. You know it can't be that simple, it never is. With systems this complex, it never will be.
So, you talk to your team. What are they feeling? You know how they are adapting, you planned that part. But you don't know how they are reacting. What you hear are vague responses about how 'things are really moving faster these days' and 'AI helps us to keep on top of the workload.' Senior staff are a bit more descriptive, but no one wants to push back against this market disruption. Can you blame them? You don't even want to push back against it. Changes are making it to the production environment and velocity is up slightly. No one is yelling extra loud or long. You've got a lot of other challenges on your plate, some pretty heavy. So, you do what any good leader would do, you grab the most dangerous one and start working on it.
The team keeps running the new model and moving forward. But you never really get a glimpse of the real problem that was already consuming them. Until, two engineers resign in the same week. One tells you they found a different role. They didn't though, you know the market is upside down right now. So, you start to suspect it was something really bad. The other one tells you they just don't want to participate in this style of software development. The reliance on AI just doesn't fit them, and they are happy to have a career change. Of course, you start the job search, but you also start the investigation internally. Eventually, one of the managers has the guts to name the real problem. There is no time for understanding.
Engineers used to spend most of their day making sure understanding was aligned and agreed upon. That day is over. The code generates too fast, and AI is being relied on to review it as well. The quote that rings in your ears from that conversation was, "if I don't have time to understand the system, how can I be held responsible for it when it breaks?" It is a great point, but it means that you don't just have an understanding problem, you have an understanding crisis.
AI is here to stay. The upside is real, and there is a lot of potential to empower people. However, engineers' ability to understand does not scale linearly. You cannot just keep adding hours of understanding time to the day and expect an equal amount of understanding to occur. There may be some gain to be had for some amount of time, but it will have diminishing returns. In addition, AI systems do not comprehend your system. At least not in the same way that engineers do.
Generating code was never the real bottleneck. Some elite teams may have come close, but I would need to see it to believe it. The real bottleneck is the time it takes to have ownership. The kind of ownership that allows you to make quick and confident decisions in order to adapt to market demand or team changes. That only comes from a solid understanding of the market, what you are building, and the team and tools doing the work. The bottom line is that no one wants to be responsible for systems that they cannot understand. That would be a form of insanity. You won't be able to predict when the next failure will be the one that costs you your job. You won't be able to protect against those failures by ensuring the quality of your products. Isn't it a bit unreasonable, and frankly, risky to operate this way?
The engineers that used to love sharing about the intricacies of your system would love nothing more than the opportunity to do it again. So, are you going to deal with the understanding crisis or wait to see whether it's a major incident or another lost team member that comes next?