DPG Media operates multiple content brands across different editorial verticals, each with its own distinct voice, tone, and audience. The challenge with deploying AI into their content pipeline wasn’t access to capable models — it was getting those models to behave as if they understood the brand. Without properly engineered system prompts, the agents defaulted to generic outputs that required heavy human editing before anything was publishable. Every agent was essentially operating without a proper brief, producing technically correct but editorially useless content that added workload rather than reducing it.

We architected a three-agent pipeline on OpenClaw Gateway with each agent assigned a tightly scoped role and a precision-engineered system prompt. The Creative Director agent receives raw inputs and transforms them into structured creative briefs — it doesn’t write content, it directs. The Brand Strategist enriches each brief with brand voice guidance, audience context, and channel-specific constraints before handing it forward. The Content Producer agent then executes against both, operating with detailed instructions covering editorial tone, sentence structure, vocabulary restrictions, and formatting rules specific to each DPG brand property.

Each system prompt was built in layers: a core role definition, non-negotiable output rules, few-shot examples of acceptable and unacceptable outputs, and explicit instructions for handling edge cases and ambiguous inputs. The prompts were tested iteratively against a benchmark set of real content briefs and refined until first-pass acceptance rates consistently exceeded 90% without human correction required.

Human revision rounds dropped by 90% compared to unstructured AI usage. Content that previously required multiple edit cycles now exits the pipeline ready to publish on the first pass in the large majority of cases. The structured three-agent architecture also gave the team a clear intervention point for each category of problem — strategic issues get addressed at the Strategist prompt level, tone issues at the Content agent level — making the system maintainable and improvable over time rather than an opaque black box that nobody knows how to fix when it drifts.