Why hasn't X given us a productivity boost?

I've heard this question many times through the years, where X is Windows, OOP, the internet, blockchain, agile, AI, etc.

Sometimes, the answer is that X is overrated. Sometimes it's that you're doing X wrong. Sometimes it's because you don't have an X-sized problem. For instance, most businesses didn't have a blockchain-sized problem.

The question I'm asked most often today is, "Why hasn't AI lived up to the hype?" There are a number of reasons. Today, I want to talk about one. And to do that, I'll focus on AI in the enterprise. And to do that, I want to first talk about agile.

I used to hear, "We adopted agile, and things got worse!"

When people tell me, "We adopted agile," what they almost always mean is, "Management forced a daily standup on the entire engineering org without bothering to get buy-in from anyone."

Management forces their team into a new process without first thinking through the second-order effects.

Let's think about that 15 minute standup. You block out 15 minutes every morning, but it really lasts 30 minutes because Bill always rambles on and on while everyone's eyes glaze over. What is the true cost of that meeting? Well, when an engineer sees a meeting on their calendar, they think, "I can't do any deep work for at least an hour ahead of that."

It takes time to get into deep work, so no one attempts it when they know they'll be interrupted by a meeting just as they're getting into their flow. Every meeting that lasts N minute probably costs you around N + 60 minutes of deep work. But there's also the post-meeting doledrums. I don't know any developer who leaves a standup energized and ready to dive into work. Instead, they take a break, reset and get into a good head-space. So, there's easily another 10-15 minute buffer following a standup. That 15 minute standup that last 30 minutes actually costs you closer to 100 minutes. Per day.

That's not exactly accurate. I do work before the standup, but I don't do deep work. I do tedious work like reviewing GitHub issues, checking email, etc. This is work that I'd probably do at some point in my day, anyway, but the standup is forcing me to do it at a time of day when I'm more capable of doing deep work-- rather than at the end of the day when my mind is better spent on shallower tasks. So, there's a real cost to that morning standup. The cost is fuzzy, and you can disagree with my numbers, but it's a real and recurring cost.

Let's be conservative and call the true cost of a standup 1 hour. That's 1/8th of your engineering hours wasted. The decision to adopt agile cost your engineering team 12% of their time budget.

A standup arguably has some value, but it's not worth 12% of your engineering costs.

But that's not the only problem. You're paying a bigger time penalty than you probably realize, but you're also paying a morale penalty. Forcing tedious, useless meetings on developers-- and in most organizations, for most of the engineers, standups are pretty useless. Maybe not 100% useless, but let's call it 80% useless. For mediocre engineering teams this isn't a huge deal. But good engineering teams will resent this. The more time-wasting bureaucratic friction you add to your engineering org, the more likely you are to nudge your best engineers towards the exit, and the more likely you are to retain the mediocre engineers.

The point is, top-down decisions without buy-in happen all the time, and have plenty of second order effects and far-reaching consequences.

What does this have to do with AI? The same management mindset that forced their teams to use "agile" forces their teams to use AI. There are huge engineering orgs that have made AI usage as part of their employee evaluations-- not using AI enough? You're on an improvement plan. What are the second-order effects of this?

Resentment, active sabotage, ridiculous workarounds and gaming the system-- all sorts of second order effects. But, productivity? No. Not because AI is a useless or unproductive tool, but because forcing AI into an org that doesn't want it, doesn't know how to use it, isn't optimized for it, and (often) actively resents it-- that will produce a lot of things, but it won't produce productivity.

Why hasn't AI given a measurable, objective performance boost to big enterprises? It's because the problems big enterprises have are not AI-shaped. They are cultural. And a broken culture is probably the hardest thing to mend.