It took me a long time to swallow this pill, but I hereby regret to inform you that the reason companies have been experiencing low AI ROI is simply the fact that employees have either A) not been delegating enough of their work to AI or B) delegating "fake" work to AI to satisfy their managers' tokenmaxxing demands.
Throughout the last few weeks, as I rotted away in two to three hour meetings, I ran the following experiment over and over again: I opened my terminal, sent a short prompt to GPT-5.5 (or Opus) explaining what the meeting was about, and asked it to solve the problem at hand. Whether it was a meeting with product managers and designers or a technical discussion about how we should implement a solution, the model returned a thorough analysis and proposed the right solutions in minutes, if not seconds.
What this made me realize is that, for most time-consuming activities, asking AI is still seen as a backup option rather than the default. We think FIRST and THEN we ask AI. And we don't ask it to solve the big, juicy problems. Instead, we ask it to solve the little issues—the tedious and mindless ones that we perceive as unworthy of our limited time. This is the single biggest reason why AI ROI has been impossible to detect for most companies who have doubled down on tokenmaxxing.
AI IS effective, and it IS making us (or at least has the potential to make us) 10x faster. But for some reason—maybe out of habit, maybe out of pride (I think it's pride)—we're not using it to fry the big fish. We're not using it to answer the big questions, the ones that require "alignment" across 3 teams and 7 layers of management.
And this makes me wonder: Do we need an entirely different corporate structure in this age of AI? In this AI era, decision making shouldn't scale linearly with team size. It should be near instant. We can no longer afford to spend months aligning on things that a model can solve in seconds. So, it's possible that our companies are too bloated and too fragmented for us to be able to leverage the tremendous power of these models.
One possible solution: large companies operate like early stage startups, embracing a messy environment with flattened hierarchies, fluid teams, and decentralized autonomy. Now that every employee has a genius, generalist LLM, we erase the boundaries between roles. Let PMs push PRs, and let engineers write PRDs. Amazon is already proving this works.
Comments URL: [https://news.ycombinator.com/item?id=48332242](https://news.ycombinator.com/item?id=48332242)
Points: 1