A Brief, a Gate and a Log: What an Agent-Era Agency Sells, in Code

In the agent-era of marketing agencies, the product sold is a concise setup consisting of a brief, a gate, and a log. This Python code illustrates this concept without referencing any specific customer story. The idea originates from public writings by BCG and Google on agent-native marketing, both emphasizing the shift towards direction and oversight by human marketers.

The first component is the brief, a set of instructions for the agent to follow. Represented as a dataclass with fields for the goal, target audience, specific phrases the agent must say or avoid, and the accountable human. The brief also includes a designated owner responsible for the work.

The second element is the review gate, a mechanism for the agent's work to be checked before publication. Another dataclass, Draft, contains the draft text, references to the original brief, and a list of flags indicating any issues found during the review. The check function scans the draft text against the phrases specified in the brief, adding flags for any violations of the "must_not_say" list or absence of phrases from the "must_say" list.

The final component is human sign-off, coupled with a log of all approved drafts. The release function enforces the check that the draft has no flags before proceeding. Upon approval, the function records the event in a log file, capturing details like the timestamp, the owner, the human who approved the draft, the draft's goal, and the learning question it aims to address. Despite its limitations, this framework captures the strategy, system design, oversight, and accountability in just 40 lines of code.