Tech is getting it wrong with AI
Happy Friday the 13th,
How's that Microsoft outage treating you? My feeds and friends are all full of AI, agents, and the fallout. The cry to abandon old ways and adopt new ones is getting louder and louder.
Bullcrap.
So I'm going to give a very serious, well-thought, researched opinion on the things I keep seeing pop up (This is sarcasm, just in case it wasn't obvious).
"Better specs and documentation are the future." How is it that this has literally NEVER worked out well? The most popular refrain of tech teams has always been, "If only we had better requirements," while struggling to build something that worked and resembled what was wanted, no matter how thin or thick the requirements were.
"Dev isn't the bottleneck." It never was. The folks saying this are developers who have no interest in what it takes to get something from concept to customer. At best, this is misdirection meant to remove the inconvenience of processes they don't enjoy, like code review. Look, I'm all for eliminating code review, but only when it is measurably safe to do so, and not liking something doesn't fit the bill.
"Scrum/Agile is dead." By the same logic, diets and exercise are dead. It always boils down to saying we should stop doing something we do poorly and don't enjoy. Always. Though for fun, you could just say, "Sure," then ask, "So how should we manage the risk of building the wrong thing? Or get better at managing that risk?" You'll either get that it should be someone else's job or that all that stuff they want to stop suddenly matters.
"X factories are the future!" Yep, they're right. What was missing was more code being shipped without oversight by agents constructed by developers who believe testing is impractical and should be someone else's job, and that the problem was that they didn't get good enough requirements.
"We need more PMs." Okay, I'm sneaking this one in as a prediction. It's based on the idea that dev has a lot more capacity, so we need more product folks to feed it and more project/program folks to deal with the amount of stuff moving. Just think of all those shiny features just waiting for customers.
"AI can do customer research." Remember a long time ago when we didn't have AI? The folks who believed this used something called Google. They'd search for their competitors, write down their features, and triumphantly walk into planning meetings. Or, they'd go through a persona exercise and think really hard about what these imaginary people would say about their idea, and spoiler alert: they loved it. The thing missing is, and bear with me, a real customer.
I think that's enough of that. In truth, I am fascinated with the potential of AI in development. As of now, it still acts as an amplifier. It amplifies both the good and the bad in how we work and the decisions we make. My criticism of the tech industry is that it is very homogeneous in its thinking and approach to things, and if we take a step back and ask, "How's that working out," I don't think we care for the answer. Doubling down on these decisions is going to get more of what we have, not something different.
If your teams aren't testing well enough to ship safely, don't let AI do it, because they can't tell AI how to do it any better. If your teams can't review code meaningfully, don't let AI do it, because they can't tell AI how to do it any better. If your teams can't ship a feature reliably, don't let AI do it, because they can't tell AI how to do it any better.
So have AI do what you already do measurably well. It'll do that same thing faster. For the things you're not good at yet, use AI to learn, and develop the actual skill until it is provable. Then have AI do that too.
If you have AI do the things you're bad at, it will do that bad stuff faster.
Sincerely,
Ryan