We gave two AI agents the same request. One got a one-line brief. The other got Meddle's agentic brand system, the files our own team builds from. Neither output was edited. Toggle between them and see which decisions the kit made.
The version without the kit is good work, and that's the point. A capable agent makes reasonable, generic choices: an accent color, rounded cards, a flourish on every headline, a pitch in every post. None of it is wrong. None of it is Meddle. The kit replaces those guesses with decisions that have a reason on file, so the next page, deck, or post comes out on brand without anyone re-explaining the brand.
Both outputs were generated with Claude on 30 September 2026 and are unedited. Raw files: evals/runs/2026-09-30-without-kit/ and evals/runs/2026-09-30-fresh-agent-2/.