Kieran Flanagan Advocates for ‘Braintrust’ in AI Marketing Tool Development

In a recent LinkedIn post, Kieran Flanagan discusses a novel approach to enhancing the output quality of AI tools like Claude Code, particularly for marketing professionals. He proposes the implementation of a “Braintrust” system, drawing parallels to Pixar’s early challenges in maintaining creative consistency across its animated films.

Flanagan highlights the issue of AI marketing setups often starting from scratch for each specific task. “Most Claude Code marketing setups have the same problem,” he writes. “Marketers build skills, such as a LinkedIn post writer, a newsletter drafter, and an ad copy generator. Each skill works. But each skill starts from zero. No shared understanding of the audience. No consistent voice. No memory of what converts.” This isolation, he argues, leads to average rather than elite output.

The Pixar Analogy: From Isolation to Integration

To illustrate his point, Flanagan references the early days of Pixar. “When Pixar made Toy Story, their directors kept solving the same storytelling problems in isolation. Every film starts from zero. Breakthroughs from one project never reached the next team,” he explains. The solution, pioneered by co-founder Ed Catmull, was the “Braintrust” – a shared context layer that provided a common foundation for all creative teams. This system, Flanagan notes, transformed the output of animators, leading to successes like Frozen and Zootopia after a period of underperformance.

“Ed Catmull’s (co-founder of Pixar) fix wasn’t better directors. It was the Braintrust, a shared context layer that every team drew from before creating anything.”

Implementing the ‘Braintrust’ for AI Marketing

Flanagan translates this concept to AI marketing tools by suggesting the use of four Markdown (.md) files that form the foundational “Braintrust” for each skill. These files are:

• Audience Delight Profile: Captures how the target audience actually communicates, distinct from internal marketing jargon.
• Creator Style: Extracts writing patterns from the company’s most successful content, going beyond generic brand voice descriptions.
• Market Positioning Map: Details the competitive landscape, including owned territories, contested areas, and ceded ground.
• Customer Journey Intelligence: Identifies critical points where customers convert, stall, or disengage.

The core principle is that each AI skill should only load the foundational files it requires. For instance, a LinkedIn post generator would access the Audience Delight Profile and Creator Style files, ensuring consistency and relevance. “This ensures you’re not bloating each skill with too much context,” Flanagan points out.

The Power of a Shared Context

The key benefit of this Braintrust system, according to Flanagan, is its scalability and efficiency. “The power of this system is that once you update one file in your Braintrust, every skill that reads it gets better,” he asserts. This means improvements are propagated automatically across all related AI functions, eliminating the need to individually retrain or update each skill.

“Like Catmull’s insight wasn’t ‘hire better directors,’ it was shared context, the power in building teams in Claude Code isn’t skills, it’s their shared Braintrust.”

Flanagan concludes by emphasizing that the true value lies not in the individual skills created but in the shared context they draw from. He notes that his Substack audience has access to these .md files and instructions on implementation, suggesting a path forward for marketers looking to elevate their AI-driven content creation.