Why AI Struggles with Content Creation Compared to Code, According to Kieran Flanagan

In a recent LinkedIn post, Kieran Flanagan explores the fundamental differences between Artificial Intelligence’s capabilities in code generation versus content creation, arguing that content remains a significantly more defensible and challenging domain.

Flanagan begins by highlighting the rapid adoption of AI in code production, noting that “Most of the early, fastest-growing tools in AI were for content creation. Years later, AI is now shipping near +60% of an org’s production code. But it’s still shipping pretty mediocre content.” He attributes this disparity to the nature of feedback loops inherent in each domain.

“The reason is obvious: with Code, you can verify whether it’s good or not; it compiles or doesn’t, the tests pass, or they fail. AI does much better when there is a binary feedback loop.”

The Subjectivity of Content vs. Objectivity of Code

According to Flanagan, code has a clear, objective measure of quality: it either works or it doesn’t. This binary feedback loop allows AI to be trained and refined effectively. Content, on the other hand, is far more subjective.

“Content is way messier. What makes good content? It’s the creator’s taste and judgment. It’s the reader’s perception of what good is. And how platforms have decided to tweak their algorithms this month,” Flanagan explains. This inherent subjectivity makes it difficult for AI to achieve consistent, high-quality output without human oversight.

Flanagan further elaborates on how AI’s patterns in content generation are more visible and scrutinizable than in code. “In code, if AI codes the same app the same way millions of times, people don’t inspect the code and say, wow, I’ve seen these exact coding patterns so many times now, because AI is generating them, so they must be slop. In content, AI uses the same writing patterns to create content; they’re on show for everyone to see, scrutinize, and get familiar with.” This leads to content that can quickly become repetitive and unoriginal, or as Flanagan puts it, end up looking like ‘slop’.

Rethinking AI Workflows for Content

The author posits that current methods of using AI for content creation are underdeveloped. He outlines several key areas where user approaches need to evolve:

• Personalized Workflows: Flanagan argues against using generic, one-to-many AI content tools. He believes that because content relies heavily on individual taste and judgment, AI workflows should be customized to the user, not standardized across a company.
• Overcoming Laziness: He criticizes the common practice of simply giving AI an idea and pasting the output, stating, “This has never worked and never will.” Effective AI content generation requires more effort and strategic input.
• Context Over Prompt: For creative tasks, Flanagan emphasizes that context is significantly more critical than the prompt itself.

“For creative tasks, the context is way more important than the prompt.”

The Ideal AI Content Setup

Flanagan envisions a more sophisticated setup for AI in content creation, involving:

• A system that learns from past successes and failures.
• Continuously updated context based on audience habits and unique content patterns.
• Skills and plugins custom-built for a specific persona, not just a platform.
• Differentiated workflows for various content lengths, noting AI is better suited for long-form content.

Ultimately, Flanagan believes AI should be used for ideation and to enhance human writing, not to replace the act of writing itself. “Used for ideation, to enhance someone’s writing, never to be used for the actual writing itself,” he states.

“I said years ago, AI is a boom for true domain experts. Never is it more obvious in content or creative tasks. AI is nowhere near good enough to replace true creative experts. It is good enough to be a great assistant, if set up correctly.”

In conclusion, Kieran Flanagan’s analysis suggests that while AI excels in domains with clear, objective feedback like coding, its application in content creation requires a more nuanced, personalized, and context-aware approach to truly leverage its potential as an assistant rather than a replacement for human creativity.