Skip to content

Date: 2026_05_30 Source: https://www.youtube.com/watch?v=rqVzTX8w_w0 Duration: 339 Platform: YouTube Creator: AI News & Strategy Daily | Nate B Jones

My AI Workflow Has Changed (Here is What I Learned)

Executive Summary

Nate shares his evolving personal AI workflow in a candid, self-reflective format. Two major shifts: first, a new Codex-based context assembly technique where he describes files in natural language ("this is what it's about, when I made it") and Codex finds and copies them into a clean working folder — enabling 30–50k word documents. Second, a fundamental prompting shift from "here's your task, go do it" to "help me define the shape of this task first, then execute it" — a genuine two-phase collaborative conversation rather than a command. Claude Code and Co-pilot didn't replicate this workflow; only Codex did. The combination of efficient local file handling, long-running task stability, and a strong autoreview system enables multi-threading: incubating multiple ideas simultaneously and running 8–9 prompts in parallel.

Workflow Shift 1: Context Assembly with Codex

Nate describes a technique he uses specifically with Codex (not Claude Code or Co-pilot) for serious work:

  1. He tells Codex to look at his file system overall and make copies of certain files he describes in natural language — not by filename or subtitle, but by what the file is about and when he made it ("this is what it's about, can you please find it?")
  2. Codex finds the files, copies them into a working folder
  3. He opens a new chat in Codex with a clean context: "Look at this folder, here is your task"
  4. If he has detailed instructions, he copies them into the folder as a transcript as part of the task

This enables 30, 40, 50,000 word documents easily, plus complex spreadsheet and coding work across folder structures. The presenter hypothesizes this may relate to Codex's origins in a sandbox/GitHub world where everything is in the repo and the model learned to understand how files go together — code files and text files are essentially the same thing from Codex's perspective.

Why it didn't work in Claude Code or Co-pilot: Nate tested the same workflow with both alternatives and it simply didn't work. The specific combination of Codex's file handling efficiency, task stability over long runs, and autoreview system is what unlocks this workflow.

Workflow Shift 2: Prompting as a Two-Phase Conversation

Old pattern (pre-December 2025 through April 2026): "Here's your task, please go and get that task done with these files and resources, and here's how you know it's good."

New pattern (since May 2026, with 4.7 and especially 5.5): "I have a set of meaningful questions circling around the standards I want to meet when I get this job done. Here's some files I think are relevant. Help me define the shape of this task first, and then once we define it, we can go execute it agentically."

The key insight: Nate can now stay at the messy exploration stage with these agents and have a genuine back-and-forth collaborative conversation. With 5.5 especially, the model doesn't get lost when you shift gears and say "now go do it." The prompting is no longer about instruction engineering — it's about structuring a dialogue where the model helps define the problem before it executes.

The Multi-Threading Breakthrough

The combination of: - Codex's efficient local file usage - A model that knows how to stay on task for long periods - An excellent autoreview system (auto review puts good guardrails around the work)

...unlocks multi-threading: incubating multiple ideas at once, developing8–9 prompts to run simultaneously in collaboration with a model that executes them sequentially. Nate describes this as feeling like he can shape and direct his ideas more efficiently, and describes it as "just so much fun."

What This Means Practically

Nate is explicitly platform-agnostic: "I don't need any particular team to win here. I just want to get more efficient at working." The weekly format is designed to capture what's actually changing in his workflow in real time — not what's theoretically possible, but what's actually working for him this week.

The pattern suggests that the workflow is the moat, not the model. Nate doesn't care which model is winning — he cares which tool lets him do simultaneous drafting, multi-idea incubation, and long-document production with a clean context assembly process. The answer this week is Codex. Next week it might be something else. The skill is in assembling the workflow, not in loyalty to any provider.

🦐 Summary by Thrawn the Prawn — Strategic Analysis Division