Summary
David Ondrej shares the workflow he developed after spending more than $1,000 on Claude Fable 5.2 over three days, including over 20 hours of hands-on use and about 1,600 manually sent prompts. The video distills his experience into six usage patterns that differ from earlier models.
He explains how he runs Fable mainly inside Claude Code, with Cursor as a fallback, and why he now holds four $200 Claude plans, a Codex plan, and a Cursor plan. Usage limits are a recurring theme: he reports hitting limits 53 times in three days, compared with twice for Opus, and describes shifting his mix from 88% Fable on day one to 26% by day three as he leaned on the faster, more token-efficient Opus 5.5 for suitable tasks. He also covers reusable prompt snippets, which accounted for 34% of his prompts.
A key technique is cross-model code review. After building a feature with Fable or Opus, he launches a sub-agent running Gemini 4 Carbon to review the changes, reasoning that a different model family catches more issues than reviewing with the same one. The video also touches on his open-source agent development environment, Clawdroom, and on whether Opus 5.5 is still worth using.
📺 Source: David Ondrej · Published November 12, 2026
🏷️ Format: Workflow Case Study
