In this talk, Mike Lyons reflects on the common mistake of automating tasks that should not exist in the first place. He shares observations from his experience at Kairos, where teams spend significant resources creating polished weekly status reports that no one reads, only to automate them and waste even more time. Lyons introduces Elon Musk’s five‑step algorithm for evaluating automation: question requirements, delete unnecessary parts, simplify and optimize, accelerate the process, and finally automate. He warns that most organizations accumulate redundant reports, approvals, and meetings that persist long after their original purpose is forgotten, and that automating such processes can amplify errors, especially when human oversight is reduced to superficial approvals. Lyons emphasizes the timeless principle that building the right thing matters more than building something right, and cautions that AI makes it cheaper to build the wrong thing, urging teams to reconsider whether a task should exist before automating it.
Chapters
You know, I talk to a lot of people and they want to know like what they can do with AI. What can they automate with AI? And I get it. It's a good question. We use AI quite a bit at Kairos. We automate everything that we can. And it's a bit kind of, it's almost addicting in some ways. But I sort of wonder about this Step earlier, and the question is whether the work needs to exist at all. I mean, over the years I've seen teams just pour real money and hours into like producing weekly status reports. They do it really fast, but and the reports you know while they're they're beautiful, like oftentimes nobody reads it. I've seen this in multiple occasions, and and somewhere in there a very capable group of people got a. Efficient at making something that nobody was wanting, nobody was using. If a task is pointless last year, it's probably still pointless this year. But now it runs on a machine at speed that we can't imagine, and you can waste a whole afternoon, you know, in four seconds. It's a kind of strange way to measure progress. It reminds me a little bit. I've got this up here on my computer in front of me. It reminds me a little bit of Elon Musk's. 5-step algorithm, but it's like a process that he refers to of like how he thinks about this. So let me just read it real quick. Number 1, question every requirement. That's a great one. Number 2, delete any part or process that you can. So this is speaking directly to like, don't automate something that shouldn't exist. Simplify and optimize, accelerate cycle time. These are— that's the 4 things. But then the last one is automate. See, it's last. First, question if it's— question the requirements, question if it should exist, simplify and optimize, delete it if you can, then do it fast, reactive fast, then you automate. That's the order. Most organizations are full of this stuff. Reports, approvals, meetings, and checkpoints that got added after something went wrong in 2014, and then they, they never pulled it back out. They never fixed it. Nobody really remembers who asked for it, but everybody keeps feeding the machine. And then there's the part where you do decide to do something about it. You do decide to automate. And, and I wonder about what's left to catch the mistake. When, when a person did it, somebody noticed when it went sideways. But when a machine does it alone, its mistake can run all night and nobody's the wiser until Monday. And the in-between version is kind of interesting too. The one where a person is supposedly checking the AI's work. And it's like, okay, you check it once, you check it twice, and then it's like, you know, it turns into clicking approve without reading. That's just kind of, you know, bad oversight. I spent a lot of years in delivery, and I remember hearing that it matters more to build the right thing than to build the thing right. I mean, yeah, you have to build the thing right, but it matters more that we're building the right thing. And that's still true. AI just makes it cheaper to build the wrong thing, which makes me wonder how much of it we should be building at all.
