Yesterday we were at an AI demo evening hosted by Worker.ai. The format was simple: live demos, one keynote, real conversation. One of the better community setups I've seen lately.
The talk that stuck: World Models
The keynote was from the team at Mimicate, and it was one of the clearest explanations I've heard of why LLMs hit a ceiling in real business contexts.
The core argument: LLMs are pattern matchers. They're excellent at generating text that sounds like analysis, but they don't actually compute anything. They confuse correlation with causality. They can describe your business situation in fluent, confident language without modeling a single variable. The phrase that stuck was calling them "excuses machines", really good at generating plausible-sounding text.
World models take a different approach. Instead of predicting the next word, they calculate the next state of a system given an action. They're constrained to physically possible outcomes, which eliminates hallucinations, and they can attribute outcomes to causes, something LLMs structurally cannot do.
The self-driving car analogy made it click. The car's AI doesn't describe what might happen. It simulates thousands of scenarios in milliseconds, picks the best path, and updates continuously as reality changes. The same logic applied to pricing strategy, inventory, or marketing spend is genuinely powerful.
The combination Mimicate is building: LLMs handle interpretation and communication, world models handle simulation and computation. The loop, hypothesize, simulate, evaluate, repeat, is what separates a system that guesses from a system that actually knows.
Worth watching closely, especially for companies making high-stakes decisions at scale.
We demoed Opmore
We showed three integrations live on the night:
- Opmore + Lovable: How company context stored in Opmore can be used to build a lead magnet in Lovable in minutes.
- Strawberry Browser + LinkedIn: How tasks in Opmore can trigger LinkedIn outreach through Strawberry Browser, keeping the workflow in one place across channels.
- Claude integration: Our own agents built into the Opmore assistant, using the company brain as live context.
On demoing
This was the first time we demoed Opmore live in front of a room. A few things became clear fast.
When you demo, you're writing prompts in real time, explaining what's happening on screen, and managing nerves at the same time. The brain is doing a lot at once. But the reaction is worth it. People engage more, ask better questions, and the impression is stronger because they're watching something work.
It also gave us a clear list of things to improve, and ideas for how to run the next one better. We'll be doing more of these.
Join the newsletter.
A weekly summary of new insights, plus what's new at Opmore. No spam, just insights.