Keep pulling the thread on Claude Code.
The speaker found Anthropic's Fable model to be exceptionally effective for his personal AI advisor skill, providing more detailed and practical advice than other models.
The US government restricted access to Anthropic's Fable model shortly after the speaker posted about its capabilities on X.
Anthropic's Fable model was able to combine a user's personal context with external research to identify business opportunities that other models could not find.
A personal AI advisor can be constructed using a system of four text files: skill.md for AI behavior, plan.md for personal context, learnings.md for storing insights, and eval.md for quality checks, which can be run in environments like Codex or Claude Code.
The speaker's personal AI advisor system, running on platforms like Codex and Claude Code, is integrated with his Mercury MCP bank account to pull financial data for tracking monetary goals.
A personal AI advisor's output quality can be improved by using an `eval.md` file, which acts as a yes/no checklist to ensure the AI's advice is grounded in personal context and actionable before being delivered.
Anthropic's Fable model identified that the primary bottleneck for growing paid subscribers on the speaker's Substack was improving renewal and retention, not new acquisitions.
Anthropic's Fable model advised that Substack Notes was an underutilized growth channel, observing that similar creators post to it multiple times a day to grow faster.