The artist using AI to sell brands | Joe Salvatore

Summary

In this episode, Joe Salvatore describes a shift from experimenting with image and video models to orchestrating multi-step AI systems for brand content production. He argues that the key scarcity is not model access or technical coding skill, but taste, direction, and the ability to write a tight creative brief. Parth Patil extends that idea into software, predicting more ephemeral, on-demand tools as inference gets faster and agents generate interfaces and workflows in real time. The discussion grounds these ideas in a concrete Ketone IQ campaign workflow that uses Hermes for research, Codex for generation, GPT Image 2 for assets, and overnight batch variation. The episode closes on Wittgenstein: AI may not “understand” like humans do, but language now increasingly summons outputs into being.

Notes

• Joe frames AI as a production layer that does not replace taste.
• Parth describes a creator support program with $1,000 a week in tokens.
• The program emphasizes tool freedom, custom agents, and closing idea-to-world latency.
• Parth identifies Joe as deeply experienced with image and video models.
• Joe says his AI curiosity started with a 2021 Lord of the Rings x Wes Anderson trailer.
• He describes four years of tool-chasing across releases and modalities.
• Midjourney was his earliest image-generation entry point.
• Joe says the workflow moved from playful sandboxing to brand content production.
• He moved from single-image prompting to group and batch generation.
• He treats Codex as a structured creative process, not just a prompt box.
• Joe points to node-based tools as a bridge to automation literacy.
• Figma Weavy is framed as a teaching tool for understanding multi-step generation.
• Joe says broader constraints usually improve creative prompting.
• He advocates high-level directives and explicit context over open-ended prompting.
• Parth contrasts objective software tasks with subjective creative evaluation.
• Joe says agents cannot tell if creative output is actually good.
• Human taste remains in the loop for evaluating creative quality.
• Joe says taste is built by repeated exposure to good references.
• He treats preproduction research as part of taste formation.
• The Ketone IQ workflow begins with Hermes-based research into decades of ad styles.
• He uses 20 historical ad styles as a research set for current sales templates.
• He combines GPT Image 2, copy generation, and template-driven ad assembly.
• He scales from a small test run to 10 variations per template overnight.
• Joe says his motivation is creative curiosity, not fear.
• Parth says the shared agent accumulates capabilities from the people it works with.
• Parth and Joe use a Telegram agent for recurring briefs and newsletters.
• The newsletter is designed to interpret the last 24 to 48 hours for future implications.
• Joe says clearer articulation is becoming a major differentiator.
• AI exposes the difference between a sloppy brief and a tight brief.
• Joe says automation clicked only after he could see the workflow visually.
• Parth says Weavy helps brands understand the path from point A to Z.
• Joe says future creative operators need to learn directing.
• Joe revisits Wittgenstein through the question of whether models really understand words.
• Reid describes AI as a set of language games with uncertain but meaningful understanding.