Summary
Nate B Jones runs an open-ended head-to-head experiment pitting OpenAI’s Codex (in Ultra mode) against Anthropic’s Claude Fable in a problem-discovery challenge: given full access to his local files and Slack workspaces, each agent must independently identify a business pain point, propose a solution, and build an automation—without being told what problem to solve.
Codex completed its run in a single pass with zero errors, auditing Nate’s media Slack and producing a research-to-scripting handoff package. The execution was clean, but the problem selected was narrow and bounded—improving an existing workflow step rather than addressing a deeper strategic bottleneck. Fable required more patience (multiple permission dialogues), but its problem identification was qualitatively stronger: it recognized that choosing which story to tell is the hardest and most painful part of a media business, and proposed a pre-pipeline idea refinement tool. Nate describes Fable’s output as immediately essential; Codex’s as merely useful.
The broader argument is that AI agents can perform autonomous problem discovery—scanning your actual behavior and communications to identify automation opportunities—without the user needing to specify the task. Jones also walks through how he packaged this into a reusable skill with configurable guardrails (blocking access to personal Slack channels, constraining research scope) for systematic deployment. The video includes context that ChatGPT Work and Codex combined are currently adding roughly one million users per day, overtaking Claude Code in total active users.
📺 Source: AI News & Strategy Daily | Nate B Jones · Published July 17, 2026
🏷️ Format: Comparison
