GTM Engineering Has a Context Problem (www.octavehq.com)

Two years into the AI wave, engineering has been reinvented because code lives in repos that encode history, tests, constraints and intent—so agents can read, run, and extend that context. GTM (marketing/sales) hasn’t made the same shift: its knowledge is “stateless,” scattered across people, Slack threads and spreadsheets, so LLMs hallucinate without a grounding reference frame. The article argues this structural deficit—not model quality—is why AI hasn’t automated GTM the way it has engineering. It maps engineering primitives to marketing equivalents: campaigns as codebase, positioning/docs as comments, A/B results as tests, and the CRM as the database. The proposed fix is a “GTM repo”: a structured, versioned store of positioning, ICPs, campaign logic, outcomes and constraints that agentic AI can both consume and help build. Practically, teams bootstrap the repo by using AI agents in a generate→critique→refine loop to extract tacit “why” and taste from senior marketers, turning tribal knowledge into executable guidelines. With integrated signals (win/loss, conversion tests, CRM data) and iterative feedback, agents can produce repeatable workflows—update positioning, draft messaging, or recommend experiments—reducing restart-from-zero cycles. The payoff is a durable moat: models will be commoditized, but teams that embed accumulated GTM context into infrastructure will compound speed and quality.