I asked Fable and Codex what my business should automate. They disagreed.

Source: Nate’s Newsletter Date: 2026-07-17 URL: https://natesnewsletter.substack.com/p/let-ai-pick-what-to-automate

Summary

Nate gave two agents (Fable and Codex) the same open brief — find what to automate in his content business — without steering them. Codex optimized a downstream workflow (research-to-script handoff) and produced the more pleasant tool; Fable identified the harder upstream problem (duplicate/weak story ideas reaching production) and, in Nate’s assessment, “saw farther into the business” despite a worse UX. He’s shipping a reusable skill that surfaces multiple evidenced automation candidates rather than picking one, deliberately keeping the human as the decider.

Implications

• Agentic engineering: a clean anecdote for a recurring finding — agent UX quality and agent judgment quality are different axes, and the more polished tool isn’t necessarily the more valuable one. Relevant to how this project evaluates agent output, not just agent ergonomics.
• Coding agents: informal head-to-head between Codex and Fable on an ambiguous, non-coding judgment task (business prioritization) rather than a benchmark — useful anecdotal signal on divergent agent “personalities,” low rigor.