> The bottleneck for small operators is different: it's not knowing which workflows are worth building, in what order, and what "system-level" vs "task-level" use actually means in practice.

Are you saying that from what you see, small operators also fail to get ROI, but for different reasons?

I work in a large enterprise. On one hand, we’re being told we should think of ways to use AI more. On the other hand, to even start (beyond just using Copilot to develop what I’m already working on), I need to have an idea and sell it to some AI board to get their blessing. At that point, I will have a microscope on me, tracking everything, to watch if this wild experiment is a success or failure. No thanks.

If they really want me to try something new, they will give me the space to try things where I am free to fail quietly and privately, pivot, and continue trying things. Asking for ship dates on day one is no way to operate projects with so many unknown unknowns. No one wants to learn and fail with an audience.

The following is my take on what's happening — outside the software-development domain, which is special vis-à-vis LLMs for obvious reasons.

Given worker access to generative LLMs, plus training and motivation to use them, LLMs are effective for certain workflows. Those workflows tend to be personal, one-offs, or summarization in nature: write a bash script for this headache I have every day; tell me what colleague X is trying to say in his 1200-word email, since his writing is garbage and he can't get to the point; "what's the Excel formula syntax for this other thing that I keep forgetting?"; etc.

So the time and mental-energy savings inures to the workers, mostly from coordination tasks that don't directly create core value. And then those savings aren't "reinvested" into value-producing activities whose benefits would inure to the firm because the workers have no incentive to do so; don't know how to create core value; don't have the skills to create core value; or aren't permitted to do those activities by higher-ups.

Bottom line: LLMs are eating busywork coordination activities — hence no impact on most firms' bottom lines.

I think you’re pointing at something real. Adoption lag matters. If the end user doesn't change behavior, ROI won’t show up no matter how much infrastructure gets built. I’d add another layer though: expectations. Many CEOs implicitly treat AI like deterministic software. install it, flip the switch, get linear productivity gains. But these systems are probabilistic. They’re "slippery" Output quality varies, edge cases multiply, and oversight is required. That makes ROI non-linear.

Exactly! having the budget isn't enough. Legacy players need to adapt processes and incentives to turn AI investment into real strategic advantage, or AI-native disruptors will outpace them.