Systematic Upheaval for Efficient Agentic Use

Whenever I see current "bearish" takes on LLMs, I look to see how the author addresses the underlying systematic changes which are bound to occur as these tools become more adopted. Typically, it's simply not addressed. I know we're well past the point of needing to highlight that "this is the worst LLMs will ever be again," etc., but there's a slightly different angle to that perspective that is arguably more important than that fact in itself: the systems we operate in now are the worst they'll ever be in terms of AI integration (there's definitely a catchier way to phrase that).

Basically, the mistake I see a lot of people making is trying to force these AI tools to work as an augmentation of current processes. Even high-tech businesses struggle with this (I've seen it first hand). And it's really something which can only be properly solved with time- you can't just upend an existing business to make it work better with AI. That's incredibly risky, difficult, and likely to not succeed. So most of us are stuck trying to make our current systems work with AI instead of trying to work within systems AI is best equipped to work in.

Almost ironically, this is the exact issue the author is pointing to (give or take). For example, AI tools won't be able to run a marketing department on their own. They can do a ton of the leg work, but without constant human guidance, none of that is useful. Sure, the AI can do some market research, but reviewing that research (especially knowing you can't "trust" it like you could a human expert) is still very costly (perhaps costly enough to offset any efficiency gains). Maybe AI can generate some content, but it won't be perfect, and the iteration on that content will still be the bottleneck (as it always is). Obviously there's plenty of value to still tap into there, but it's nowhere near as transformative as needed to match "the hype" of AI.

But imagine a new company, created today, by people who don't have much to lose. Their goal is to be AI-first. That research? Ship it. The AI is good enough. That content generation? Build a harness around the process and take the human out of the loop as much as possible. It doesn't have to be perfect- if it's fast and cheap enough, then it may be valuable enough. And once you start doing stuff like this, you can start incorporating *new* processes which these other firms simply can't fit into their workflows. Maybe AI can beat these other firms to the punch by actively monitoring social media, maybe AI can flood the market in such a way that these older firms just can't break through, maybe AI can produce something that is so odd that it captures attention very effectively, etc. And this new firm is operating in a way that actually allows this to occur. There's not much to risk (relatively speaking), so why not?

But it's not just that; it's also leveraging emerging systems which are designed around AI, too. Perhaps another new firm comes up with a digital billboard system which focuses on AI-first integration. Not an existing billboard company who cheaply plugs an MCP in the hopes of looking modern; but an actual "AI-first" system which can match the pace of other AI-first firms. Now that AI-first marketing firm can take advantage of new aspects of the system which older firms simply can't tap into. Things move fast, novelty encourages more novelty, and all of this is happening *way* faster with *way* less overhead.

Of course, all of this is just as speculative as suggesting AI will get so smart that we'll enter a utopia soon so none of this matters (or whatever your take is). I'm not going to pretend to be able to predict what that future may look like. But I *do* know these systematic changes take a long time unfurl, and we're far from seeing any sort of end-state relative to AI technologies. I think much of the "hype" (or the bubble, or whatever) around AI is actually based on *this* premise, not on the idea that AI will be a drop-in replacement for labor in our current systems. I think these companies, investors, etc. all recognize and understand this dynamic and will continue to pump these systems full of money to accommodate these inevitable changes. And it will come relatively slowly, but it will speed up once successful examples are cracked and firms will be more willing to take on these more dramatic shifts in how they work.

Note: I actually like that blog post I linked to at the top of this post. It is a refreshingly grounded and non-hysterical "bearish on LLMs" take, so I'll give the author credit for presenting their arguments effectively without resorting to the kinds of assumptions, hypotheticals, emotional appeals, etc., typically employed with these kinds of posts.