Descriptions:
Nate B Jones shares a candid update on how his personal AI workflows have shifted in recent weeks, centering on a specific Codex technique for managing context windows via the local file system. Rather than manually tracking files by name or path, Jones describes telling Codex to find relevant documents using natural language descriptions — “this is what it’s about, this is roughly when I made it” — and having the agent copy matching files into a clean working folder. A fresh chat is then opened pointing at that folder, giving the model a well-structured context window without the noise of an entire file system. He reports being able to handle 30,000–50,000-word document tasks reliably with this approach, as well as complex spreadsheet and coding work.
Jones explicitly notes that the same workflow does not transfer to Claude Code or Claude’s co-working mode, positioning this as a Codex-specific capability tied to how the agent was trained on repository-style file structures. He also traces how his prompting philosophy has evolved across three distinct phases: pre-2025 prompt engineering for structure; early 2026 task-delegation with evals; and a current phase — enabled by Claude 4.7, Claude 5.5, and a refreshed Codex — where he collaborates with the model to define the shape of a task before shifting into autonomous execution.
The video is useful for practitioners looking to understand how leading AI power users are adapting their workflows as long-context agentic models mature, particularly around multi-threading simultaneous drafting tasks and using models as co-designers rather than pure executors.
📺 Source: AI News & Strategy Daily | Nate B Jones · Published May 30, 2026
🏷️ Format: Workflow Case Study
