I think the judgement angle is the only interesting part of this article, and the piece worth pursuing is automating the judgement where possible.

Note: I am assuming that it is 2027-28 and reliable AI automated coders exist (or the equivalent workhorse AI in your field), which makes implementation time negligible compared to making decisions. The effect is somewhat weaker with present-day-level AIs. I'm also assuming that the 100-person company is very competent with AI outside of making decisions, but that the startup can plan things much faster due to not needing a committee to do so.

Very rough maths:

If your 100 person team still follows collaborative processes to cancel out errors (let's say it takes 10 people a day to decide on a single deliverable's shape), then give the design to the AI to implement (as we assume the AI can do it without supervision), then you can ship 10 deliverables a day.

At the same time, that 4 people team can have all of them bouncing ideas off of AIs to help them make decisions in rapidfire all day. They'll each individually spend an hour working on a decision then hand it to an AI. Their decisions are on average as good as your 10-member team meetings because while your medium-sized company's decisions sometimes end up suboptimal due to politics, the startup's decisions are individuals so make the wrong call more often, and I assume these two effects cancel out. In that case, your competitor with 4 people cranks out 32 deliverables a day assuming that the implementation AIs don't have to be supervised at all.

In summary it's not "move fast and break things", it's just "move fast, focus on making decisions, delegate everything else to the AI". Remember that the decisions are all that matters if the AI can do all the implementation.

Mmm, thats a lot of assumptions that all have to hold true to make the math work, like, you're starting to venture into hiring 9 women to make a baby in a month territory here.

But it also makes some more fundamental assumptions that I'm trying to challenge a bit. It assumes delivering 32 "deliverables" per day (the meaning of which is context specific) is better than 10. Is that always true? Is that delta the most relevant factor in the success of a business compared to its competitors? Etc.

I agree, quality of decisions will be much better than quantity. I'm trying to keep decision quality constant here

A lot of the above assumptions are just to keep things fair since in the real world there are a lot of variables that can't be ignored. For example, keeping AI competence equal between the two companies.

I'm just trying to show that under my assumptions (~2027-28 AI, highly competent) it is quite conceivable that a 4-person visionary company can start a beat a much larger traditional one on quantity, not that it will definitely happen. I guess it's even rougher than the "rough math" I said it is.

I guess the point I'm trying to make is this: startups have always been able to beat big companies in serial execution speeds, but beating them in straight-up parallel work quantity is very unusual, but I think there's a good chance it will happen. This is simply because decision-making scales really poorly in traditional companies by headcount and I think it'll get more and more important relative to implementation work. Hence the focus on quantity of "deliverables" (i really mean medium-size project designs, think equivalent to 5-day targets for a dev team which I assume to be AI's average task horizon before it needs human decision input)

At least, the small team will win for some time until we get straight-up superintelligence that replaces the decision-makers as well. At which point the calculus suddenly flips and the richest company wins by default.

Yeah, I don't necessarily disagree assuming those assumptions end up being true. It just wasn't really the point I was trying to make. More about the mentality that you need to keep up with the proverbial Joneses and even sacrifice other things in order to do so, or else you will be left behind. You see this on threads where people are talking about their own personal development as well. Fears about being left behind, or people warning others that if they don't go all-in on agentic development they will be left behind. It just feels so oversimplified, almost philosophically empty.

I can understand fears about employment of course, people need money to live and support their loved ones, but beyond this there is more to life than just slopping out more and more code that "does things" and it's worth considering if there is any real benefit to do so, or even if the negative aspects outweigh the positive ones. I suppose we will find out in real time if everyone's side projects are better to be left unfinished, or finished by vibe coding. It's not so clear to me.

I'm not advocating for not utilising LLM's in companies or whatever, just that prioritising velocity at the expense of everything else might not be so valuable. Of course if there are no downsides than it would be purely valuable, but that basically begs the question lol.