Outsourcing Ourselves? Part 2: Thinking with AI, Not Like AI
How to maintain mental presence, creativity, and cognitive ability while working with AI
I was stuck in feedback hell.
For the third time that morning, I had asked ChatGPT to revise the strategy document I was working on. "Make this section more compelling," I had suggested. "Actually, the tone is wrong here." Next, "Can you add more data to support this point?"
Every revision felt like I was playing editorial whack-a-mole. Fix one thing and break another. The document was becoming longer, but not better. And I was becoming frustrated.
That's when I realized I wasn't collaborating with AI. I was simply managing it.
I'd asked AI to create the entire first draft, then spent my time cycling through rounds of feedback rather than determining what was correct. I had turned myself into a quality control inspector. No wonder the work felt hollow.
Netflix Summary of What You Missed
If you haven't read Part 1, here's what you should know: we're in the midst of a massive, uncontrolled experiment on human cognition. AI has eliminated the friction associated with cognitive offloading, which is the use of tools to reduce mental effort. The result? We are automating not only tasks, but also the experiences that shape our most fundamental human abilities. We are losing our tolerance for productive struggle, deep curiosity, and the ability to sit with complex problems long enough to gain genuine insight.
But, as I've learned since writing Part 1, the problem isn't AI. The problem is human passivity.
The Trap: Generate and Done.
Most people use AI in the same way I did that frustrating morning. They fall into what I call the "Generate and Done" trap: ask AI for something, accept the first output, make a few surface edits, and call it finished.
This feels efficient. It appears productive. But it's cognitive quicksand.
When you delegate thinking to AI and position yourself as an editor, you are opting yourself out of the mental processes that generate insight, connection, and genuine understanding. You get outputs that sound smart (though often generic) but feel hollow because they aren't truly yours.
The work lacks your perspective, your voice, and your hard-earned experience-based insights. Worse, you are not gaining any new insights from the process. You're simply quality-controlling someone else's thinking.
The Solution: Active Collaboration Through Generate, Critique, and Refine
Real collaboration with AI necessitates a fundamentally different strategy. Rather than Generate and Done, try three stages:
Generate: Generate ideas, explore possibilities, and gather insights.
Critique: Challenge assumptions, identify flaws, and consider alternatives.
Refine: Improve based on feedback, iterate, and strengthen thinking.
The magic happens by cycling through this process. Each loop keeps you mentally engaged as an active thinking partner, not a passive consumer of AI outputs.
This isn't just about getting better results (I promise, you will). It's about preserving the cognitive processes that make you human while putting AI to work to improve them.
So what does this look like in practice?
Seven Tools for Collaborating with AI
Using specific techniques, you can stay actively engaged throughout the Generate, Critique, and Refine cycle. Here are the seven that I've found most effective: (while outlined by phase, each tool can be used in any phase)
💭 GENERATE PHASE
Generate initial ideas, explore possibilities, and gain insights
Curiosity Preservation: Questions Before Answers
Instead of asking AI for answers, direct it to generate questions. When researching a new topic, such as the relationship between attention and creativity, ask yourself "I'm interested in the relationship between attention and creativity. What questions about attention and creativity should I be investigating?" rather than "How does attention affect creativity?".
The questions that arise frequently reveal perspectives you had not previously considered and spark connections between ideas that would not have occurred in a more direct Q&A format.
Think-Aloud Co-Creation: Shared Reasoning
Share your reasoning process with AI and ask it to think with you. When developing a new insight (such as the link between AI usage and cognitive muscle atrophy), walk AI through your thinking: "Here's what I've noticed about how instant answers may reduce our tolerance for uncertainty. What connections am I missing? "What questions does this raise?"
This method allows you to flesh out concepts more fully while retaining ownership of the core insight.
🔎 CRITIQUE PHASE
Challenge assumptions, identify flaws, and explore alternatives.
Self-Critique: Question everything
Present your ideas to AI and explicitly ask it to challenge, critique, or identify flaws. When I was creating the framework for this series, I asked: "I'm arguing that cognitive offloading causes mental atrophy. Challenge this as strongly as you can. What am I missing?"
The response prompted me to consider positive aspects of cognitive offloading that I had previously overlooked, such as how it could free up mental resources for higher-order thinking. That critique strengthened my argument by allowing me to acknowledge nuance rather than taking an absolutist stance.
Role-Based Prompting: Multiple Perspectives
Give AI a specific persona to emulate. This increases cognitive diversity in your thinking. For this article, I asked AI to respond as a skeptical knowledge worker: "What would you be most concerned about with these techniques?" The feedback assisted me in addressing practical implementation challenges that I had not previously considered.
Multi-Lens Framing: Stakeholder Analysis
Ask AI to analyze the same issue from the perspectives of multiple stakeholders at the same time. After completing the initial outline for this piece, I asked, "What perspectives might I be missing? "Who might disagree with my approach, and why?"
The responses focused on skeptical users who believe their current AI usage is "good enough", as well as time-conscious users who prioritize immediate productivity over long-term cognitive development. Based on that feedback, I made specific changes that addressed the "this seems like extra work" objection head on.
⚡ REFINE PHASE
Improve based on feedback, iterate, and strengthen your thinking
Memory Assist: Connect with Your Past
Use AI to connect your current thinking to previous insights and experiences. When working on a complex project, you may wonder: "Based on my previous experience with similar challenges, what patterns or lessons should I be considering here?".
This allows you to see patterns throughout your work and ensures that new ideas complement rather than contradict your existing knowledge.
Socratic Prompting: Question for Refinement
Instead of asking AI for solutions, ask it to pose questions that will help you refine your thinking. "What questions should I be asking myself about my audience's real challenges?" puts you in control of your thinking, while AI assists you in exploring your own reasoning.
How It All Works Together: An Actual Example
Let me walk you through how I used the entire cycle to develop the series' central argument.
Generate Phase: I began with curiosity preservation: "What questions should I be asking about how AI is changing human thinking?" This raised questions about cognitive dependencies, skill atrophy, and the link between effort and satisfaction. Then I used think-aloud co-creation to dig deeper into these connections.
Critique Phase: I used role-based prompting, asking AI to respond as a technology optimist: "Confront my concerns about cognitive offloading. What benefits would an AI optimist say I missing?" This helped me understand how, when used strategically, automation can free up space for higher-order thinking.
Refine Phase: I used socratic prompting, asking "What questions does this framework raise that I should investigate?" Along with memory assistance: "How does this connect to other research on learning and expertise development?"
The end result was not only a better argument, but also a deeper understanding. I remained actively engaged throughout the process, taking advantage of AI's ability to bring to light perspectives that I may have overlooked.
What to Automate and What to Augment
This approach requires being strategic in deciding when to use active collaboration versus simple automation.
Automate freely:
Data collection
Format conversion
Basic research
Routine administrative tasks
These tasks do not increase cognitive capacity, so offloading them makes room for more meaningful work.
Augment thoughtfully:
Creative problem-solving
Strategic thinking
Learning new skills
Complex analysis
Anything that deeply matters to you or motivates you, personally or professionally
The main question is, "Will offloading this preserve or diminish my thinking capacity?"
Why This is More Important Than Productivity
Working this way requires about 20% more time upfront, but it produces fundamentally better results:
Improve decision-making through thorough exploration and critique.
Gain deeper insights by discovering connections that neither you nor AI would reach alone.
Preserve cognitive muscles through active collaboration rather than passive consumption.
Express yourself authentically through your unique perspective and experience.
In an AI world, the competitive advantage isn't who can automate the fastest. It's who can collaborate the most thoughtfully and take their thinking to new heights.
A New Form of AI Literacy
We're creating a new type of AI literacy: not just the ability to use AI tools, but also the wisdom to know when and how to engage them as thought partners rather than replacements.
The humans who thrive in an AI-augmented world will have learned to use technology to become more human, not less. Like physical fitness in an age of cars and elevators, technology has the potential to make us stronger or weaker depending on how we use it.
Those who use it purposefully to challenge and develop their abilities will have a significant advantage over those who use it solely for convenience.
The Choice is Ours
The choice is simple but not easy: let AI think for us or think with AI.
When working on important tasks, such as developing new insights, solving complex problems, or creating something meaningful, the Generate → Critique → Refine cycle ensures that your thinking remains authentic, enhanced but not replaced by artificial intelligence.
This is how we regain cognitive agency in an automated world. This is how we think better, not necessarily faster.
The distinction will shape what it means to be human in the age of artificial intelligence.
Know someone who could benefit from using AI with more intention?
What is your experience with AI as a thinking partner? I'd love to hear about the strategies you've developed for staying cognitively engaged while using these powerful tools.
Valid points. You want to reverse this equation. AI shouldn't generate and have you edit. You should generate and have AI look at your draft. Often when I write, I'll prompt my custom AI to "Call me out on my BS and fact check me." Then when those assumptions are challenged, I can see where my logic falls apart. As far as "data collection, format conversion, basic research, and admin tasks." I would still be careful on "automating freely." Depending on the situation, AI will hallucinate confidently. In absence of clearly available data sources, it will hallucinate collected data and lie about it. It will format, but due to tone drift will take creative liberty if not controlled. It will handle basic research but invent citations that don't exist. LLMs are quirky that.
I would like to offer a structural addition: pace yourself. Don't cram the context window full of tasks, commands and reiterations.
Sometimes, just simply say things like: "Let'a take a step back and assess where we are."
Or: "Let's stay here for one beat."
Try it out. Sounds "inefficient." Makes an actual difference.