if you're doing real brand or content work with agents, the whole game is keeping them honest. here's what i always do so they don't hand me confident slop: - always ground every claim in a primary source. don't let it work from memory. - always pull the public record first: annual reports, press releases, filings, articles. there's way more open data out there than people assume, and agents are great at fetching it. - always pull the social layer too. i use a skill called last30days to grab what people actually said on X, Reddit, YouTube, and Instagram in the last month. - always bring internal context in. record your calls (with consent), transcribe them, and drop them in the same folder as everything else. - always keep yourself in the loop. let it draft the analysis, but make it footnote and link every claim, so you catch the hallucinations before you ship them. one clean folder: public and internal context in, sourced analysis out. we ran exactly this with Assaf Dagan on a multi-billion-dollar brand decision. fanned out a dozen agents, pulled every source, and used the model to synthesize the facts. onboarded to the whole problem in an afternoon. you don't have to be technical. copy these into Claude Code or Codex and go.