✱ Button 2026 · Content design conference · Speaker

Less vibes,
more structure.

The mechanics behind content design for AI agents: router logic that directs intent, structured output that keeps agents on-brand, and rubrics that turn "this feels off" into data.

Router logicStructured outputEval rubricsDev-to-English
Watch the talk ↗ Watch copy write itself →
Less vibes, more structure. The mechanics behind content design for AI agents. Amit Singh, Button 2026.

Customer“Why are you not paying my California taxes?”

California tax setup is NOT_READY. Missing: State Employer Account Number. Signed POA required.

Looks like your California tax setup isn't fully connected yet. I know this is frustrating when you expect things to just work. We're missing your CA EDD account number. It works like a routing number for California tax. Do you have it on hand? I can walk you through adding it right now.

No one wrote either of these. The craft didn't disappear. It moved.

Seven parts of an agent. Content design has a job in every one.

The response is the box we've always owned, and the smallest on the page. Hover a part.

InstructionsPersona, reading level, never invent rates, educate don't advise, never a dead end. The first artifact that's unambiguously ours.
  1. Model01
  2. Instructions02
  3. Router03
  4. Tools04
  5. Memory05
  6. Guardrails06
  7. Evals07
“A chatbot talks. An agent decides.”
“A skill is not a second brain.”
“Design the data, not the sentence.”
“You cannot ship a stated behavior that is not tested.”
The blueprintUnderstand→Design→Influence→Measure