Summary
Jeff Su shares six AI productivity techniques designed to stay useful as models change, organized into three goals: wasting fewer tokens, doing less manual work, and getting AI to learn your preferences over time. The tips apply to ChatGPT, Claude and Gemini.
The first two tips focus on saving tokens. The “markdown shortcut” converts text-heavy PDFs into lightweight markdown files; in his example, a 661 KB PDF shrinks to a 94 KB file the AI can process faster. He also recommends giving AI plain-text formats like CSV rather than messy spreadsheets, and storing originals as a fallback referenced in an agents.md or claude.md instructions file. The “folder map” tip has the AI build an index.md describing each file in a folder so it only opens what’s relevant, with an optional scheduled task to keep the index current.
Later tips include reviewing old saved instructions, since rules written for earlier models can hold back newer ones, and the “intent check,” where the AI repeats back your goal and proposes a plan before starting high-stakes work. Su demonstrates it with a wedding-agency comparison, where the plan step produced a rubric-based evaluation rather than a bare recommendation. The video ends with a bonus lightning round.
📺 Source: Jeff Su · Published November 10, 2026
🏷️ Format: Tutorial Demo
