How the term is used in this materialIn the channel's material, system prompt shows up in the context of Val Town's AI assistant Townie. Your prompt is combined with what founder Steve Krouse calls a big ass system prompt explaining how Val Town works, and both go to the underlying model, Claude 3.5 Sonnet [^ebfee90b]. Why it made the differenceKrouse credits the system prompt as one of three reasons Townie rebuilt a full-stack podcast app from a single prompt in under a minute — a task that took roughly an hour a year earlier: - Platform facts: the prompt tells the model that code runs in Deno JavaScript, not Node or client-side JS, and that it never needs to start or deploy a server. - Bug avoidance: it documents which platform bugs to steer clear of. - Documented goodies: platform features — like the OpenAI integration used by a generated app's question button — are written into the prompt so the model can use them directly, replacing docs a human would otherwise paste in [^ebfee90b]. Why it mattersThe system prompt is what turns raw model capability into a working product: it encodes the platform's rules and features so the model can produce correct code on the first try. It's also a maintained asset — Val Town iterates on prompting per model, and even rolled back a model update when they couldn't get its prompting right [^ebfee90b]. - A system prompt is fixed instructions sent alongside the user's prompt to shape the model's behavior. - Val Town's prompt teaches the model how Val Town works: Deno JavaScript, no server management, known bugs to avoid. - Documented platform features in the prompt let the model use them without pasted docs. - Prompting is per-model and iterated on — even rolled back when a new model couldn't be prompted correctly. [^ebfee90b]: Do 60-Minute Coding Tasks in 60 Seconds—With AI — Ep. 41 with Steve Krouse.