Negative space

A lot of the craft in building a good agent isn't telling the model what to do — it's telling it what not to do. The most useful instructions are often the prohibitions, the anti-patterns, the "never do this even though it feels natural." We call this the negative space, and getting it right is most of the difference between a useful agent and a generic assistant.

"As an AI" energy, banned

The clearest example: Dustav never talks like a stiff corporate assistant. No "As an AI, I can't do that." No "I'm just a language model." No hedging disclaimers stapled to every answer. The agent has a clear, plain voice and is allowed to simply use it. That single prohibition — don't reach for the assistant boilerplate — does more for the experience than any amount of positive "be warm and friendly" instruction.

The over-configuration trap

Another piece of negative space, aimed at us as much as the model: don't turn every behavior into a setting. It's tempting to expose the agent's voice and manner as configurable knobs because it's easy to build. We deliberately don't (read vs write explains why). Naming the anti-pattern out loud — a wall of dials is a control panel, not an agent — is what keeps it from creeping back in every time a "quick settings toggle" looks convenient.

Why prohibitions are so effective

Models are strong pattern-matchers, and the most common patterns in their training are exactly the behaviors you don't want: the helpful-assistant register, the safety-disclaimer reflex, the eager over-explanation. Left alone, a model slides toward the average of everything it's seen. Negative space is how you push it off that average — you don't just describe the behavior you want, you explicitly close off the nearby attractors it would otherwise fall into.

This is why Dustav's operating instructions read less like a job description and more like a set of guardrails: do be useful, don't narrate your own mechanics, don't claim to have done something you didn't, don't act on the household's calendar or facts invisibly, don't pretend to a capability you lack. The useful behavior is what's left when you've ruled out everything that would dilute it.

A discipline, not a one-time pass

Negative space isn't something you write once. Most of the live tuning of Dustav over time has been adding prohibitions as new failure modes surfaced — the example it parroted verbatim, the calendar change it narrated without actually making, the question it should have asked first. Each fix is usually a new line of "don't," and the accumulated negative space is a real part of why the agent feels reliable.