Former Google productivity expert Jeff Su details the transition from 'AI Literate' to 'AI Native' by implementing specific habits like text expanders and reverse prompting. He argues that while AI agents are the future, current success relies on human ideation paired with rigorous AI execution workflows.
Overview
In this deep-dive interview, Jeff Su, a former Google employee turned full-time creator, dismantles the fear that AI will replace creative jobs, arguing instead that it will replace those who refuse to integrate it. Su introduces a maturity model for AI usage: moving from 'Curious' (sporadic use) to 'Literate' (paid subscriptions and prompt libraries) to 'Native' (integrated habits and friction-free workflows). He shares tactical frameworks, including the 'Reverse Prompt' technique to extract reusable templates from successful chats and 'Red Teaming' to stress-test business offers. The conversation explores the practicalities of scaling a content business, the reality of 'AI Agents' being a decade away from perfection, and the sheer grit required to bridge the gap between a corporate career and full-time entrepreneurship.
Key Points
The Hierarchy of AI Competence: Su categorizes users into three levels: 'AI Curious' (sporadic, free users), 'AI Literate' (paid users who understand model differences and save prompts), and 'AI Native' (users who build habits to reduce friction). Being 'Native' involves integrating AI into the OS layer using text expanders and designing workflows that assume an AI collaborator exists from day one. Why it matters: Moving to the 'Native' level drastically reduces the cognitive load and friction of using AI, allowing for compound productivity gains rather than one-off wins. Evidence: AI native, I would say you're developing and maintaining AI native habits. And for example, I talk about this a lot. Number one, using text expanders.
The Reverse Prompt Technique: Instead of settling for a good output after a long back-and-forth conversation, Su recommends asking the AI to 'reverse engineer' the entire chat into a single, optimized prompt. This converts a one-time success into a repeatable asset that can be stored in a database for future use. Why it matters: This eliminates the need to master prompt engineering from scratch; it allows the user to leverage the AI's understanding of its own logic to create perfect templates. Evidence: You end the conversation with, hey reverse engineer a conversation and write the single prompt that would have produced my final response in one go.
Iterative Learning via Prompt Improvement: Beginners should stop using their raw initial prompts. Instead, they should ask the AI to 'optimize this prompt' for a specific model (e.g., Claude 3.5 or GPT-4). Crucially, the user must read the refined prompt to understand why it is better (e.g., structure, context, persona), thereby upskilling themselves through observation. Why it matters: This acts as a continuous feedback loop, training the human to think like the model over time. Evidence: Ask AI to improve your prompt. Read through the prompt. Do this 10 times a day. You're going to be better than 99% of people.
Red Teaming Your Own Work: Su utilizes a 'Red Team' prompt where he asks the AI to assume the role of a skeptical customer or critic to find blind spots in a script, offer, or idea. This simulates an adversarial review process to identify weaknesses before going public. Why it matters: It prevents echo-chamber thinking and allows creators to pre-emptively address objections, making the final product significantly more robust. Evidence: Assume the role of my ideal customer persona and tell me what your concerns are or why you would not buy my product... that's the power of the red team technique.
The Reality of AI Agents: Despite the hype, Su aligns with Andrej Karpathy’s view that we are entering a 'decade of agents,' meaning reliable autonomous agents are not fully ready for consumers. True agents require three things: reasoning, access to tools, and the ability to iterate/refine. Current security risks (like prompt injection) make them dangerous for handling sensitive data like emails. Why it matters: Understanding this timeline prevents businesses from over-investing in immature technology while preparing them for the eventual shift to autonomous workflows. Evidence: But more recently, Andrew Kaparthy came out and said it's a decade of AI agents... meant implies there are a lot of technical and behavioral issues... before there's widespread adoption.
Sections
Strategic Observations
Meta-level synthesis of Jeff Su's approach to AI and content creation.
The 'Curse of Knowledge' is the primary barrier to AI adoption; experts forget that beginners lack the vocabulary to even ask the right questions, making 'AI Native' habits (like text expanders) invisible to the 'AI Curious'.
Prompt engineering is shifting from a creative writing task to an asset management task. The value isn't just in writing the prompt, but in cataloging, retrieving, and deploying it via system-level tools (Alfred/Raycast).
The 'AI Bubble' in finance is actually a net positive for infrastructure. Similar to the dot-com burst leaving behind fiber optics, the AI burst will leave behind massive compute and energy infrastructure that will power the next era.
Human 'Ideation' remains the only unassailable moat. While execution (editing, scripting, data analysis) can be fully augmented, the initial spark and personality injection must remain strictly human to maintain connection.
Implementation Plan
Concrete steps to move from AI Literate to AI Native.
Install a text expander tool (Alfred for Mac, Raycast, or Beef Text for Windows) to deploy frequent prompts instantly.
Create a 'Prompts Database' (in Notion, Apple Notes, or Google Keep) to store successful prompts, rather than relying on chat history search.
Implement the 'Reverse Engineer' workflow: At the end of every successful AI session, ask it to generate the single prompt that would have created that result.
Set up 'Evergreen' custom instructions in ChatGPT that define tone (no hype, no emojis) rather than defining your specific job role (which leads to pigeonholing).
Map out the content creation workflow and identify 'Red Team' insertion points to stress-test scripts and offers before production.
Verbatim Highlights
Memorable quotes regarding AI philosophy and tactics.
If you're not using AI tools right now you should be worried about creators who have embedded AI into their workflows. They're just going to make better content, funnier content, more engaging content, cheaper and faster.
Ask AI to improve your prompt. Read through the prompt. Do this 10 times a day. You're going to be better than 99% of people.
Reverse engineer a conversation and write the single prompt that would have produced my final response in one go.
I basically convinced myself... I was like, if I'm able to upload one video every week for two years, I've won.
You want to save information where you're going to use it, not where you found it.