AI and Agents

AI and Agents is the category for the agent itself: what it is made of, what it may touch, and what it takes to keep one running inside a marketing team. The pillar piece, planned first, will ask why so many marketing AI pilots stall before production, and whether the author's own answer, the marketing harness (skills, tools, context, measurement, and controls around the model), holds up as the thing to build instead.

The pieces after it will take the harness apart. One will cover agent skills, the written procedures an agent loads when a task calls for them. One will cover guardrails for agents that spend money: autonomy tiers, spend caps, and who owns the write path. The connector pieces will ask how an agent should reach a marketing tool at all, through MCP, an API, an SDK, or a command line, what each costs to maintain, and will keep a scorecard of what each ad platform's connector can read, write, and spend. A piece on tool design will ask whether a few task-shaped tools beat many raw endpoints.

Two pieces are for teams with engineers: when to build a marketing MCP server of your own, and what to check before trusting a connector you did not write. One is for agencies and holding companies: running many brands on shared infrastructure without one brand's data reaching another's agent.

The reading order below starts with the harness, which the rest will refer back to.