[Me & X] New practices in vibe coding
X : What are the new practices in vibe coding, AI agents and agentic AI?
Me : Those practices are still emerging and many of the candidate practices are still based upon the idea that the output will be deterministic code (which I view as flawed). I've only got hundreds of hours of vibing experiments (not thousands) under my belt, so some more work is needed before I feel comfortable with running a population study and talking about those practices. I'm many years away from doing this and probably won't run the study until 2030.
X : The practices will have established by then.
Me : If you say so, I don't. The practices look likely to be fundamentally different (with a heavy biological theme) in the same way that we had a radical change that my population study picked up in 2011 (based upon the preparation work done in previous years). Back then practices changed from N+1 to design for failure, disaster recovery to chaos engines, scale up to scale out, change control to continuous deployment.
I have to be thorough here, because when I run those population studies then I usually get a barrage of "nonsense", "gibberish" and "no clue" for the following 3-5 years and the usual suspects waiting on the change to fail so they can claim "mapping is rubbish". Of course, that same crowd tends to slink off into the shadows a few years after that and almost never admits they were wrong. One or two exceptions were kind enough to apologise and admit their mistakes in 2011 after giving me earache. We're still not there for the 2021 study (I've attached both).
So, I will publish something when I'm ready, confident and have run a population study. It'll take time though.
X : How long do these practices take to become normal?
Me : You mean "accepted" as a new norm, that doesn't mean everyone does them though. That takes a lot longer. To become accepted, 7-9 years, hence I run the population study every ten years as I'm a glutton for punishment.
X : How do you know these studies will be right?
Me : I don't. They are best effort based upon research and a population study usually of several thousand companies. I use maps to inform this all. But maps are imperfect. If they turn out wrong, I have no problem saying they are wrong. So far, they've held up quite well.
X : AI is happening faster.
Me : If you say so.
X : I disagree with your 2021 version.
Me : Doesn't surprise me. Come back 2028-2030 and see if you still disagree. By 2031, I'll have a new list you can disagree with.
X : Do you know there are books being written on those new practices?
Me : I know people are jumping the gun and getting on a bandwagon. I would advise against it but that's upto them. They'll just release future versions which will change the practices, trying to match what is happening. What they will skate over is the damage caused by companies adopting practices which are figments of imagination and not grounded. It's like that bimodal nonsense. C'est la vie.
Originally published on LinkedIn.