Spiral: Designing an AI Ghostwriter With Taste
Timestamps are as accurate as they can be but may be slightly off. We encourage you to listen to the full context.
Podcast Summary
In this episode of AI & I, host Dan Shipper sits down with Danny Aziz, General Manager of Spiral, to explore the philosophy and craft of building AI writing tools that enhance rather than replace human thinking. (05:26) The conversation reveals how Danny completely rebuilt Spiral from the ground up over seven months, transforming it from a simple text transformation tool into an AI writing partner with editorial taste.
- Main Theme: The episode centers on using AI as a craftsman's tool - leveraging technology to amplify human creativity and thinking rather than generating mindless content slop.
Speakers
Danny Aziz
Danny Aziz is the General Manager of Spiral, Every's AI writing partner, and a solo engineer who rebuilt the entire product from scratch using AI development tools. He combines a craftsman's philosophy of "measure twice, cut once" with cutting-edge AI development, having become one of the top users of AI coding tools like Droid and Claude Code to create beautifully designed, intentional software products.
Dan Shipper
Dan Shipper is the host of AI & I and co-founder of Every, a publication focused on AI and business strategy. He's an experienced entrepreneur and writer who has built multiple AI products and regularly explores the intersection of technology and human creativity, bringing deep insights into how AI tools can enhance rather than replace human judgment.
Key Takeaways
Slow Down to Speed Up: The Power of Intentional AI Writing
Most AI writing tools prioritize speed over quality, leading to generic output that requires extensive manual editing. (08:29) Danny discovered that the best AI writing actually comes from slowing down the process - using reasoning models that think for several seconds and showing users the AI's complete thought process. Rather than generating instant responses, Spiral asks probing questions first, creating a collaborative interview process that helps users clarify their actual intent. This approach transforms AI from a content slot machine into a thoughtful writing partner that helps users think better, not just write faster.
Multi-Agent Systems Beat Single Large Models
Building effective AI writing tools requires multiple specialized agents rather than one large model trying to do everything. (13:41) Danny learned this through trial and error - attempting to have one model handle both interviewing and writing led to context rot and degraded performance as functionality was layered on. The solution was creating a handoff system where an interviewer agent gathers information and context, then passes the complete conversation to a specialized writer agent. This architectural approach allows each agent to excel at its specific task while maintaining the full context window, resulting in dramatically better output quality.
Show Your Thinking: Transparency Builds Better AI Partners
The best AI tools make their reasoning process visible by default, not hidden behind optional expandable sections. (11:02) Danny made Spiral's thinking process open by default because true collaboration requires understanding your partner's perspective. When users can see how the AI is reasoning through a problem, they can quickly identify when it misunderstands something or approaches the task from an unhelpful angle. This transparency enables more effective feedback and course correction, turning the interaction into a genuine partnership rather than a black box that occasionally produces useful output.
Good Writing Flows From Good Thinking
The fundamental insight driving Spiral's development is that good writing is downstream from good thinking, not just better prompting. (33:16) Danny realized that most AI writing tools encourage users to outsource their thinking entirely to the machine, which produces generic content that lacks authentic voice and insight. Instead of optimizing for maximum content output, Spiral focuses on helping users clarify and develop their ideas through thoughtful questions and multiple perspective exploration. This philosophy positions AI as a thinking partner that amplifies human creativity rather than replacing it with algorithmic content generation.
Craftmanship in the Age of AI: Maintaining Standards While Leveraging Tools
You can use AI tools extensively without compromising on quality if you maintain clear standards for the end result. (40:49) Danny demonstrates this by building an entire sophisticated product as a solo engineer using AI coding tools like Droid, while never accepting subpar outputs. The key is developing an internal sense of what "feels right" - whether in code architecture, user interface design, or content quality - and consistently applying those standards regardless of whether AI or humans generated the initial output. This requires letting go of ego about who writes the code while maintaining rigorous judgment about whether the result serves the intended purpose effectively.
Statistics & Facts
- Danny became one of the top users of Droid AI coding tool, consuming so many tokens that he exceeded the maximum plan limits before the billing period ended. (39:34) This demonstrates the intensive use of AI development tools by serious practitioners building complex products.
- The DSPy-optimized judging prompt achieved 76.5% accuracy in matching Dan's editorial preferences, compared to only 58.8% accuracy with a human-written Claude prompt. (52:34) This shows how automated prompt optimization can significantly outperform human intuition in creating consistent AI behavior.
- Claude Opus 4.0 became the first model capable of reliably judging writing quality without rubber-stamping, consistently providing boolean answers about content quality that matched human judgment. (45:30) Previous models would typically rate all content as "B+" regardless of actual quality.
