The unfortunate implication of running a newsletter is that the content has to be at least news-adjacent. And sadly, the articles making the news are centered on LLMs — which I am tired of writing about, and you are probably tired of reading about.

For those of us familiar with pace layers, language models (or as they are called in marketing lingo, artificial intelligence) manifest solidly in the “fashion” layer. AI is used because it feels cool; it is the sports car of productized technologies. The underlying tech may be ”here to stay” but as soon as it falls out of vogue, it will “stay” in users’ garages, because trotting it out would be less of a status symbol and more of an embarrassment.

Five separate people, unprompted, have sent me this excellent essay on how the present-day coolness of AI has captured executives in an AI prisoner’s dilemma in the past week. The entire essay is worth your time; honestly, if you are pressed for time, I would click on no other link in this issue and re-read this one twice. Nikhil’s piece nails the current crisis of decision-making perfectly: regardless of what you think about AI, you cannot argue that it is a novel technology, and is therefore subject to all the concomitant risks on top of the ordinary risks of any software project.

Many industry leaders (misled by the sports car aura of LLM tech) have forgotten that AI is, after all, just ordinary software. Last week, the first automated reasoning program celebrated its 70th birthday. Bragging about your AI-native org is about as impressive as bragging about a transistor radio native org. Or a Xerox photocopier native org. In other words: you were already behind, you were always behind, and any commercial wins you have achieved in the meantime have been despite your technological maturity rather than because of it. This is worth thinking about, but I’m not going to do it for free.

So how do you run a software project effectively? What a coincidence that you should ask! I had just this conversation with Jason Knight on his podcast the other day.

If you don’t want to take my qualitative findings as data, Peter Merholz has some quant-based insights for you: only execs are excited about AI, and everyone else is already exhausted by the hype — because it does not match up to reality.

As far as executives are concerned, one of the major draws of AI has always been exculpation. Whatever they have provided (or failed to provide) to their employees no longer matters; AI is pitched as a great reset that is meant to wipe away all other material conditions. Part of the all-permeating hype of the AI industry is that every org is just full of ideas ready to burst forth and LLMs are the mechanism that will somehow effortlessly unlock them.

To any experienced professional, this concept should be laughable on its face. It is self-evidently not true. And yet it continues to echo in AI marketing, possibly because that is the only way this trillion-dollar bubble can continue to justify itself.

Cory Doctorow has a deep dive on why the core value proposition of AI (increasing throughput by maximizing productivity) is fundamentally flawed: that formulation has only ever been a top-down mandate, and top-down mandates always result in a degradation of quality. This is because computer applications do not eliminate work; rather, they transfer it from managers (who buy the products) to their reports.

(As an aside to myself: Dean Peters’s comment reminds me to do a piece sometime on the value of speculative fiction to product management practice.)

This issue doesn’t have a neat conclusion where I tell you how things are getting better. The trends we see today will continue until organizations feel like stoking the AI boiler burns more fuel than it is worth. In the meantime, the most important work you can do may be to look after yourself, and manage your own burnout.

— Pavel at the Product Picnic