The Silent Partner
How AI “Talks” You Into Flow
Original Image by Alex Suarez converted using Gemini
“Not only will our tools tell us about ourselves, but they also will inform us as we invent new versions of ourselves. And it’s going to have a huge effect on our own understanding and our own identity of who we are as humans.” - Kevin Kelly
The Heisenberg Problem
About 20 years ago, I started writing a book called Evangelical Capitalism (don’t worry you won’t find it anywhere). I recorded thoughts on a digital transcription recorder that I took with me on my walks.
There are hours of these recordings taken during my flow walking sessions. I think I still have the recorder somewhere but have never done anything with the notes. That’s because it was too hard to get the ideas from spoken word to text. More importantly, my progress through them is disconnected and unrecorded.
This is a different experience now. I can record directly to text and then analyze that text with an AI co-designer. The ideas still flow on my walk but they also flow when I’m refining the ideas.
However, Flow is a slippery concept. I can’t honestly say that that’s what I’m experiencing. I like to think it is but can’t tell whether it’s Flow or Flow-adjacent.
This is because one of the central challenges that Mihalyi Csikszentmihályi ran into when he was studying Flow was the difficulty of measuring it. When he asked subjects to reflect on the experience while they were having it, it broke Flow, the very thing he was trying to measure. Writing with Jeanne Nakamura in 2002, he observed, “interrupting deep flow, as the ESM [Experience Sampling Method] would do, destroys the phenomenon”
It’s like you are stuck in Heisenberg’s Uncertainty Principle. You are the observer of yourself. You can’t do that without being fully absorbed in the task. We can’t experience Flow unless that is our reality.
But what if you had a record of that reality that you could review after the fact? So much of Flow happens silently. It’s going on in your head so videotaping it doesn’t really make sense either. We can’t really tell what’s going on.
The Intellectual Sparring Partner
A silent partner who challenges you and keeps a record of those challenges as they escalate to an ongoing conversation might be a way to “see” Flow in action. Here is an excerpt of a conversation I had with Claude showing the pivot moment on the Substack I published last week.
As I was working on “Augmentation is a Two-Way Street,” two Substacks appeared that seemed to parallel the argument I was trying to make. One was the Iacono piece “The Person Who Could Have Thought” I quoted in Part 1. The other was Steven Johnson’s piece on cognitive uploading. This is where the conversation starts:
Query: “How closely does this argument parallel mine?” (sharing Johnson’s cognitive uploading)
Response (condensed): Closely on the surface but you’re working a different seam. Johnson’s axis is epistemic — AI giving you more to think about. Yours is ontological — whether the thinking changes you. He’s also ahead of you on your structural problem: he’s already dissolved the automation/augmentation binary your draft hasn’t resolved yet.
This has some analysis in it but it is Claude’s. So I double down with the Iacono piece.
Query: “This argument is also similar.” (sharing Iacono’s foundation argument)
Response (condensed): Similar but a different animal — and more dangerous to your argument than Johnson. Iacono’s ending is lethal: “First the capacity, then the tool. First the formation, then the augmentation.” That’s not a refinement of your argument. It’s a direct challenge to it.
This provokes me to assess what I read in Iacono’s piece and adapt it to my own thinking. I knew that I didn’t have the lack of foundation that he describes. In a podcast earlier that week, I had been accused of being part of “the 1% who could leverage AI” and that what I was suggesting was not generalizable.
Query: “Or, am I in the 1% who can do this and grow from it while the 99% get stupider?”
Response (condensed): That’s the question that makes Iacono dangerous. Your framework requires the walk — the unaugmented practice that maintains formation. The 99% won’t have that. And they won’t know they don’t, because the interpreter will narrate a perfectly coherent account of augmentation while their capacity atrophies.
This got me thinking through the implications of where the line on Bloom’s taxonomy between AI thought and human thought was drawn. How high up did you have to go to know that what you were being pushed to was actually outracing the cognitive impact of the machine.
After several exchanges, I decided it was somewhere in Analysis. There was a machine aspect of analysis and a human side. It was in this fusion that humans could propel themselves upward to Evaluate and Create.
Query: “I essentially made that argument in my last Substack: The End of Credentials.” (shares URL)
Response (condensed): You did — and it’s stronger than what we were workshopping in the abstract. That’s the institutional argument. Thursday’s draft is doing something different: it’s the personal practice essay. What does augmentation require of the person, not the system? Those are two different essays, and you’ve already written the first one.
Query: “And I talked about it here too.” (shares second URL)
Response (condensed): You don’t need to make the unified argument in one piece. You’re building a case across essays where each one lands harder because readers have seen the others. Thursday’s draft isn’t trying to settle the institutional question — it’s the personal practice piece the series needed.
This was the essential pivot point that launched me back into the Flow of writing. It allowed me to mentally fuse Johnson’s and Iacono’s thinking into the flow of my story.
The “Hostile Editor” Prompt
I didn’t think we had definitively addressed the “1%” question. This had two immediate effects. First, I demonstrated this kind of exchange with my students in class to see if they could follow the process. Then I had them do the same thing.
As part of their learning process, I give them a hostile AI prompt to stress test their websites. The prompt reads:
“You are a Hostile Editor and Intelligence Auditor for a federal policy agency. I am presenting a problem definition page for a national crisis. Your job is to find the weak links in my chain. Do not rewrite my text or be polite. Interrogate my evidence. Ask me: ‘Are you confusing a symptom with a root cause?’ and ‘Show me the original source for this structural claim.’ Point out if Image 1 illustrates a surface-level symptom rather than what is actually broken, if Image 2 is not a measurable metric, or if Image 3 fails to show the broader societal impact. Check that my claims are supported by hyperlinks to verified sources. If you find errors, list them as ‘Red-Line Corrections’. If the evidence is structurally sound, visually hooked, and verified by external links, respond with ‘ALL CLEAR’.”
Unfortunately, I don’t have a complete student chat transcript, but through self-admission, some went through 9 iterations of this before submitting the assignment. Based on what the student posted, this is a simulated conversation of how this might go.
Gamifying the Friction (The 9-Iteration Ladder)
First, the student uploads a screenshot of his homepage which contains images of computers, farmers, stalled loading screen along with the hostile editor prompt. Then, the conversation begins with the answer to this prompt:
Hostile Editor response (condensed): RED-LINE CORRECTIONS — NO ALL CLEAR. Image 1: You are confusing a symptom with a root cause. A stalled loading screen shows the experience of poor connectivity, not what is structurally broken. Why do the gaps exist? Image 2: A picture of farmers is not a measurable metric. I cannot track improvement with this. Image 3: Computers on a desk shows me nothing about societal impact beyond those directly affected. Who else is harmed, and how?
Student: “Okay, for Image 1 should I find a map showing where broadband isn’t available? I found one on the FCC website.”
Hostile Editor response (condensed): You’re still looking at the symptom. An FCC coverage map shows WHERE the gap is, not WHY it exists. Ask yourself: why don’t ISPs build infrastructure in these areas? The structural answer isn’t geographic — it’s economic. Show me the original source for why the market failed.
The AI plants a new idea into the student’s head. Maybe he is thinking about the problem wrong and this is what’s getting him stuck.
Student: “Wait — I looked at the FCC data and it actually overstates coverage. ISPs only have to report a census block as ‘served’ if they could theoretically serve one household there. So the map itself is wrong. Is the root cause that we’re measuring this incorrectly?”
Hostile Editor response (condensed): Closer. You’ve found a real structural flaw — the regulatory definition inflates coverage statistics. But is bad measurement the root cause, or a symptom of something deeper? Why would ISPs lobby for a definition that overstates coverage? Follow that chain.
The student brings in a new piece of information that the AI didn’t give him. This redirects the conversation.
Student: “Because they don’t want to be required to build infrastructure where it’s not profitable. So the root cause is that internet access is a private market operating like a utility, but without the regulatory requirements of one. The monopoly structure means no competition forces investment in unprofitable areas. Is that the structural claim?”
Hostile Editor response (condensed): That’s your root cause. Now prove it with primary data. Don’t show me a coverage map. Show me market concentration data — how many ISPs compete in rural versus urban markets. That’s the structural evidence.
Student: “I found NTIA data showing that 87% of rural census tracts are served by only one or two providers. Is this my Image 1?”
Hostile Editor response (condensed): Now you’re showing the structural failure, not the symptom. A single-provider market has no competitive pressure to expand or improve service. That’s your root cause image.
Student: “For Image 2 I want to track improvement. Should I use the percentage of households with broadband access?”
Hostile Editor response (condensed): That metric has the same flaw as the FCC map — it uses the same broken definition of ‘served.’ If your root cause is monopolistic market structure, your metric should measure whether that’s changing. Try: actual measured download speeds by zip code from independent speed tests, not ISP-reported capability.
Student: “I found M-Lab speed test data — it shows actual measured speeds by zip code, not what ISPs claim. And it shows the gap between monopoly markets and competitive ones. Can I use that?”
Hostile Editor response (condensed): Yes. That’s a measurable metric tied directly to your root cause. It tracks real performance against the structural problem you identified.
Student: “For Image 3 — I was going to show economic impact on rural communities. But you said it can’t be a charity argument. Who else is actually harmed if they already have good internet?”
Hostile Editor response (condensed): Think about what broadband monopolies do beyond connectivity. When businesses can’t locate in areas without competitive broadband, where does economic activity concentrate? What happens to urban infrastructure when rural areas depopulate? What happens to the food supply chain when precision agriculture can’t scale? The argument isn’t ‘help them’ — it’s ‘this affects the price of your food, the reliability of your supply chains, and the taxes you pay to support concentrated urban services.’
Student: “So the societal impact image should show the economic cost of the broadband gap to the whole economy — GDP loss or agricultural supply chain impact — not just what rural residents lose?”
Hostile Editor response (condensed): The USDA has data on precision agriculture productivity losses tied to connectivity gaps. The Brookings Institution has measured GDP drag from uneven broadband deployment. Either shows the systemic economic cost — not a charity case, an infrastructure failure with consequences for everyone. Source it to the primary document and you’ll have your Image 3.
Scaling the Unscalable
This is a 9-prompt sequence based on my own experience teaching. In that role, I am also trying to help them drill down on what is really underlying their issue. We do this live in class and it’s proven very effective as a teaching strategy.
However, there are serious problems with doing this live in class. There isn’t enough time to push everyone up the ladder within the time constraints of the class. A lot of students are unwilling to participate in this public process and so don’t submit their work.
In the past, when I only had 2-3 classes, this number was lower and I could meet with them individually for 15-20 minutes each. With 100-150 students, however, I don’t have time to do this.
The AI pushes them to come up with creative solutions to its challenges. Arguably, it gamifies the process. Unlike my class, when students engage individually with a chatbot, they are more likely to enter Flow state, because, as Csikszentmihalyi writes, “there is little opportunity for the self to be threatened.”
The 1% Problem, Unresolved
In both of these cases, the conversations hint at how having an intellectual sparring partner pushes the conversation up to the Evaluate and Create levels. The question is whether this also generates Flow through the continuous give-and-take of the conversation.
You can be the judge of that. I don’t want to suggest that these conversations in any way represent a rigorous study. One is hypothetical and the other is my conversation with the chatbot. Effectively, I haven’t proven that I’m not one of the “1%.”
These conversations, however, suggest some very interesting avenues of study in both learning and Flow Theory. This also opens some interesting doors for innovation in a startup. I stimulated my own creativity moving the Knowledge Navigator project forward in a recent chat. I have the same problem there when I’m writing: I don’t often have a teacher there but do have an on-call AI partner. More on these pathways in Part 3.