What is AI Psychosis?

Mitchell Hashimoto posted a recent tweet

https://xcancel.com/mitchellh/status/2055380239711457578


I lived through the great MTBF vs MTTR (mean-time-between-failure vs. mean-time-to-recovery) reckoning of infrastructure during the transition to cloud and cloud automation. All those arguments are rearing their ugly heads again but now its... the whole software development industry (maybe the whole world, really).

It's frightening, because the psychosis folks operate under an almost absolute "MTTR is all you need" mentality: "its fine to ship bugs because the agents will fix them so quickly and at a scale humans can't do!" We learned in infrastructure that MTTR is great but you can't yeet resilient systems entirely.

The main issue is I don't even know how to bring this up to people I know personally, because bringing this topic up leads to immediately dismissals like "no no, it has full test coverage" or "bug reports are going down" or something, which just don't paint the whole picture.

We already learned this lesson once in infrastructure: you can automate yourself into a very resilient catastrophe machine. Systems can appear healthy by local metrics while globally becoming incomprehensible. Bug reports can go down while latent risk explodes. Test coverage can rise while semantic understanding falls. Changes happens so fast that nobody notices the underlying architecture decaying.

I worry.```

To me a big question in the AI world is - what changes?  Has my mental map adjusted and in what ways?  And maybe I see the more important question as - what stays the same.

People love to ask this question - "What happens when (code is free)[https://jamesin.substack.com/p/what-happens-when-code-becomes-free]?". 

But ironically if you think code has never been the bottleneck, as I don't, then the answer is.  Not much.  This is going to be a remarkably unfocused attempt to focus, but it makes me want to circle around my understanding of leadership.  Maybe the project management side, and think about, what are the principles I carry with me?  And how does AI fit in.  How should I be thinking about the problem?

There are thought leaders at my company pushing everyone to use AI.  And ironically, despite my hesitation, I think this is the right call.  Maybe this is a first principal, but systems should be adversarial.  You don't want an environment where 20 engineers all agree on how much AI. You want a few engineers just going all out.  Maybe a few engineers a little more bearish.

But then, just because I agree someone in the company should be all out bullish, doesn't mean that I should just do what they say, in fact just the opposite, and what I want to try and do is balance various principles or ideas that I carry with me as a manager.

1. Goldratt's theory of constraints suggests that every system has at least one constraint (or bottleneck) that limit's it's ability to achieve it's goal, and that organizational performance is limited by this factor.  What this means practically, is that icing AI over everything won't have a substantial impact on productivity unless it hits the bottleneck.  I think this means as a leader, my primary goal should be to figure out what the bottleneck is and hit it.
 
2. A strategy is an area of strength against an area of weakness.  The bottlneck is the area of weakness.  How you attack the area of weakness depends on the skills of your organization.

3. You need 20% time.  You won't get to know how to use AI unless you're able to experiment in a distributed way.  Everyone should be using and testing AI and most of these experiments won't move the needle because (1)

4.  AI's most significant benefit comes from improving DevEx.  It is good when engineers are happy.  It is good when CI/CD takes less time.  AI allows us to attack all the little problems, bugs and tech debt that we didn't have time to before.  Take advantage of it.

5. We probably shouldn't abandon existing best practices because AI.  Atomic changes, automatic rollback, automated testing.  These practices exist for a reason.

6. I think the definition of Psychosis is using AI in a way that doesn't either (1) Get your most critical customer or revenue generating work done, (2) Attack the bottleneck or (3) Save Money or (4) Make everyone's lives easier.

The purpose of this article is to I think continuously remind myself that the most important thing you can do as a leader is 1: Identify the strategy / most important thing to work on, both internally and externally.  Externally it's what thing unlocks the most revenue.  Internall it's, what is the bottleneck that will unlock the most productivity.  If you're beating around the bush, and not looking for bottlenecks, then you're not improving your organization.