The Real Reason Your Agents Produce Generic Output
Most operators chasing GTM automation are asking the wrong question. They want to know which agent to deploy next. The better question: what does that agent actually have to work with?
Jared Robin's Human Tokens surfaces a pattern that any honest RevOps practitioner will recognize immediately. Zapier's GTM lead Lindsay Rothlisberger built what she calls a shared context layer—a living brief covering brand voice, ICP nuance, and deal patterns—and made it the input every GTM agent draws from before doing anything. Without it, she found agent output read as generic noise. With it, agents produced work that sounded like someone who actually knew the business.
This isn't a prompt engineering trick. It's an architecture decision.
What a Context Layer Actually Contains
A context layer is not your CRM. It's not a folder of old decks. It's a deliberately maintained document (or set of documents) that captures the judgment calls your best reps make automatically:
- ICP nuance: not just firmographic filters, but the behavioral signals and situational cues that actually predict a good fit
- Deal patterns: what kinds of champions close, what objections correlate with late-stage churn, which segments have a shorter sales cycle and why
- Brand voice with teeth: not