5 reasons LLMs might have actually peaked the way we know them Like any new tech that actually works - internet, blockchain, tulips (just to be a bit cheeky) - the excitement seems slightly exaggerated. Here are a couple things I noticed and my takeaway. 1. Robots.txt. Basically, platforms tend to protect their data. Even those relying on generating LLM traffic like reddit and LinkedIn limit "scraping" requests. The likes of X.com have fully blocked any scraping and push users to X API with extortionate prices. 2. Paywalls. Anything remotely valuable is hidden behind paywalls. Authentic thoughts that used to be blog posts are now substack subscriptions. And I didn't even start talking about data 3. "Poisoned pills". The amount of "It's not Y, it's Z" copy on the internet makes me want to become violent. New gen of general LLMs eats what old gen of LLMs has produced 4. Cost. Ok, the search market is conquered. Now losing money doesn't really make sense. Anthropic already hit hard with it's OpenClaw decision. I've seen reports of people going up to 50x on actual cost. Because guess what: AI requires compute power. And it's expensive. Especially in the geopolitical energy crisis. 5. Ancient "automate what works" problem. I remember our first CRM integration in a bank I used to work at in 2008. We didn't really have a defined sales process. And I mean - not even remote. No meeting reports, no performance metrics, just end of month revenue assessment. And then suddenly everyone had to fill in a form with 20 inputs after each call. Obviously, didn't fly. And that's what is happening now. Processes that don't work (and people have very vague understanding of how they should) suddenly get an AI wrapper. And that's all it is. A wrapper. LLMs don't create value. They help cut time. That's how they should be referred to. A time saving tool. They don't invent. They don't create. They use the available input to generate an output that requires checking and cleaning. When we launched Alfred, our AI assistant, the effort to prevent hallucinations was probably 80% of the work. My key takeaway: I strongly believe all those AI agents will generate evidence that will lead to the only possible conclusion: LLMs are a wrapper. And the hype will go down. Real adoption will bring real people in. Policies and standards will be created. Vertical clear use cases generating 1 outcome at a time. Just like a good software. Security, reliability and compliance will take over again (they always do when any process settles down). And the winners? Data and infrastructure holders.