Earlier AI models would praise any writing given to them. A breakthrough occurred when the Spiral team found Claude 4 Opus could reliably judge writing quality, even its own. This capability enables building AI products with built-in feedback loops for self-improvement and developing taste.
Related Insights
Users mistakenly evaluate AI tools based on the quality of the first output. However, since 90% of the work is iterative, the superior tool is the one that handles a high volume of refinement prompts most effectively, not the one with the best initial result.
Unlike previous models that frequently failed, Opus 4.5 allows for a fluid, uninterrupted coding process. The AI can build complex applications from a simple prompt and autonomously fix its own errors, representing a significant leap in capability and reliability for developers.
Instead of manually refining a complex prompt, create a process where an AI agent evaluates its own output. By providing a framework for self-critique, including quantitative scores and qualitative reasoning, the AI can iteratively enhance its own system instructions and achieve a much stronger result.
The frontier of AI training is moving beyond humans ranking model outputs (RLHF). Now, high-skilled experts create detailed success criteria (like rubrics or unit tests), which an AI then uses to provide feedback to the main model at scale, a process called RLAIF.
Spiral's redesign was driven by the principle that "good writing is downstream of good thinking." Instead of just generating content, the tool focuses on helping users explore and clarify their own ideas through an interactive, question-based process, making the AI a partner in thought.
A key strategy for labs like Anthropic is automating AI research itself. By building models that can perform the tasks of AI researchers, they aim to create a feedback loop that dramatically accelerates the pace of innovation.
The recent leap in AI coding isn't solely from a more powerful base model. The true innovation is a product layer that enables agent-like behavior: the system constantly evaluates and refines its own output, leading to far more complex and complete results than the LLM could achieve alone.
Most AI writing tools produce generic content. Spiral was rebuilt to act as a partner. It first interviews the user to understand their thoughts and taste, helping them think more deeply before generating drafts. This collaborative process avoids "slop" and leads to more authentic writing.
To codify a specific person's "taste" in writing, the team fed the DSPy framework a dataset of tweets with thumbs up/down ratings and explanations. DSPy then optimized a prompt that created an AI "judge" capable of evaluating new content with 76.5% accuracy against that person's preferences.
The best AI models are trained on data that reflects deep, subjective qualities—not just simple criteria. This "taste" is a key differentiator, influencing everything from code generation to creative writing, and is shaped by the values of the frontier lab.
