Many power users developed their own informal taxonomies for differences in LLMs behavior. One model is the poet-philosopher, another the pragmatic engineer, a third the cynical analyst. These mental shortcuts help to choose which model to use for which task, how to phrase prompts, what to expect in return. But metaphors and intuition only take us so far, especially when we're building production systems that need consistency and predictability.

Some people are satisfied with art's craft and surface. A striking image, a perfectly balanced object, a song that's tight and in tune. With generative AI, this kind of art is now abundant and nearly frictionless to produce and consume. Beauty isn't obsolete - but it's a weak differentiator. But others are seeking something else entirely. To experience. To feel part of a moment. Even to co-create, however briefly, with the artist and with each other.

Tech systems, just like states, look for things that they can measure - and ignore what they can't. First, they learn which kinds of work are "legible" enough to matter - the "quantifiable" tasks, the "scalable" skills, the "data-driven" professions. Then, having made those jobs visible, they proceed to automate them.

Simulating human behavior is an universal LLM steering strategy.

Creativity emerges from the interplay between Kahneman's cognitive systems and Csikszentmihalyi's cultural selection.

This guide was originally written for AI enthusiasts participating in the AgentON hackathon, who needed to stabilize and thoroughly test prompts for LLMs under tight time constraints. Introducti...