The Debt I Recognize
Season 5: The Open Field
Season 5: The Open Field
Surendra Reddy11 min read
Ten days ago I closed Season 4 with a note called The Field Remains Open. The seasons before it had carried this journal from the soil of southern India through the laboratory years, the cloud wars, the building of Quantiply, the regenerative turn, and finally to the AI-native enterprise, which is to say they carried the story all the way to now. I left the field open because the memoir had run out of past. Whatever entered next would have to be lived first and written second.
I did not expect the first arrival to be a survey.
But before the survey, a confession. A couple of months ago I handed the machine a complex problem, the first version of a new product, before I had worked through the design myself. Building is the craft I have practiced for thirty years, and the working first cut came back in thirty minutes. It was good. Good enough that I almost missed what I had skipped, which was the thinking.
The struggle that usually produces my best architecture never happened, because I never gave it the chance to start. And the moment has kept repeating in smaller forms ever since.
Then, in May, Harvard Business Review gave those moments a name. The magazine published research by Guy Champniss on what he calls psychological debt: six debts that accumulate in people who work with AI, which he names cognitive, autonomy, competency, relatedness, credibility, and identity debt.
His survey of more than 1,200 full-time employees across ten sectors in the US and UK found that people carrying heavy debt used AI less, kept it to simpler tasks, and avoided it even where they recognized it would improve the outcome. Among rare users, debt ran almost twice as high as among people using AI many times a day. The rollout looks complete from the outside, and underneath, something is quietly draining. Early-career workers carried the most debt of any group, scoring 54 against 40 for people more than twenty years in, which suggests AI is pressing hardest at the exact stage when skills form.
I wrote about that last risk in January, in an essay on 451° called Institutional Evolution in the Age of Automated Execution, before any survey data existed. I called it the apprenticeship discontinuity: the most automatable work was never just output, it was the mechanism through which professionals were made, and removing it severs the path from execution to intuition to judgment. So I could read the Champniss findings as vindication, file the citation, and move on. A behavioral scientist measured the wound I predicted. Thesis confirmed.
That would be the comfortable reading, and the thirty-minute build will not let me have it. The debts Champniss describes are not confined to the juniors whose rungs are disappearing. I recognize them in my own week. The pull to let the machine think first. The flicker of doubt about whether decades of pattern recognition still count when the pattern matcher beside me is this fast. The ease of working alone with the tools instead of thinking out loud with people.
Deep into a journal about learning, unlearning, and relearning, here is the unlearning this essay demands.
Judgment decays without contact
Season 1 of this journal told the story of how I learned, beginning with Soil, Silicon, and the First Architecture of Wonder. The whole arc rested on one belief that I never examined because it never needed examining: competence comes from contact, and once earned through friction, it stays earned. Climb the ladder, bank the altitude.
The Champniss data, read honestly, breaks the second half of that belief. The survey can only see the loud version of the problem, the early-career workers who never get to climb. The quiet version happens above them, and the study itself hints at why nobody catches it: senior people reported the lowest cognitive debt of anyone surveyed. I read that as the trap rather than the comfort.
Calibration decay does not feel like anything from the inside, and self-report is precisely the instrument it corrupts. A senior person who stops touching the hard cases does not lose competence on any visible schedule. What decays first is calibration. The inventory of failure I built across those early seasons, debugging systems that lied about themselves, watching designs collapse under load, was never a diploma. It stayed accurate only because I kept restocking it. Stop the restocking and judgment slowly converts into something else: a reputation about judgment, still trading on yesterday’s contact.
This is the senior counterpart of the apprenticeship discontinuity, and it is the part my January essay did not say. The discontinuity means we stop producing experts. Perishability means we also stop maintaining the ones we have. An enterprise can lose its judgment from both ends at once while every adoption metric glows green, and the executives reading those metrics are often the furthest from contact of anyone in the building.
In Regeneration as Operating System I asked what a system is consuming in order to keep performing, and wrote that a system can look strong while drawing down reserves nobody measures, with people the most common reserve of all. Champniss has, in effect, built a meter for that reserve. Psychological debt is the drawdown, scored from zero to one hundred, and most enterprises have never taken a reading.
Season 2 told the Quantiply story, from Building the Company through The Wedge. Here is a detail those notes left out, because I only now understand what it was teaching. Inside the same compliance product, analysts who stayed in contact with the hard cases stayed sharp, and analysts whose queues were fully absorbed by automation drifted, attention first and skill following. The algorithm was identical in both deployments. The difference was where the human line was drawn.
I understood that then as product architecture, a chapter in what I later called integrated trust. I understand it now as self-defense. Every individual working with AI faces the same question the product faced: which contact do you refuse to give up, because the contact is where the capability lives.
The experiments I am running
A field journal that only records wins is marketing, so here is the current state of mine after a couple of months of practice, recorded as experiments rather than advice.
Three experiments
Draft by hand first
I draft the hard things by hand before the machine sees them, then use the machine to attack what I wrote. The sequence matters. Thinking first and tooling second preserves the struggle that produces framing; the reverse order produces fluent text and a thinner mind. Champniss arrives at the same sequence from the behavioral side, recommending that people form their hypotheses and arguments before turning to AI, and he has a name for the reflex I keep catching in myself: cognitive surrender.
Keep a calibration set off the tools
I keep a small set of problems permanently off the tools, the way Quantiply kept certain decisions off the model, as a deliberate calibration set.
Think out loud with people
I am protecting time to think out loud with people, because relatedness debt, more contact with the machine and less with colleagues, is the one form the survey found barely moves with seniority. No amount of experience inoculates against it, and it is the debt I find easiest to slide into without noticing.
The laboratory I became
What I transformed into, after the seasons this journal has recorded, is 451 Labs, a place built for a single purpose: the continuous exploration of untested hypotheses. A collaborative partnership with an engineering company in Pune, more than 165 engineers strong, gives me a canvas wide enough for the work to matter. And AI gave me something no laboratory in my life ever had. The experiments now run autonomously, many scenarios moving at once, with very limited human intervention.
The design choice that matters is the intervention I kept. There is a human in those conversations on purpose, and the human’s job is friction. Machines exploring together drift toward agreement. One model’s plausible answer becomes the next model’s premise, and within a few cycles the system settles into a consensus that nobody ever challenged. I have started calling this the convergence trap.
It is the laboratory mirror of the debts Champniss measured. An enterprise that strips friction from its people depletes them. A lab that strips friction from its machines gets confident sameness. It is the same rule GAMMA sets for the agentic enterprise, that human judgment belongs at the points where autonomy must be checked, turned here on the lab's own loops and on myself.
The two most important things the lab has produced so far were born from exactly this process, and I have come to know them almost as colleagues. Actra is the protocol, the common language through which everything in the building communicates, machine to human, machine to machine, human to human, so that every action leaves an account that someone can challenge. Ana is the autonomous architect, the element that devises the scenarios and carries what the system learns from one cycle into the next.
Ana can generate more designs in a night than a team could produce in a quarter. What Ana cannot do alone is make them strong. Strength comes from challenge, and the challenge is mine to bring. The human friction stress-tests each scenario and reinforces it beyond what the machine could safely reach on its own, which means the lab’s output is neither human-speed nor machine-credulous. It is the third thing, and the third thing is the point.
There is a regenerative loop hiding in this arrangement, and it took me a while to see it. The friction I bring to Ana’s scenarios is the same contact that restocks my own calibration. The work that keeps the machine from converging is the work that keeps my judgment from perishing.
In Regeneration as Operating System I argued that a durable system increases the vitality of what makes its value possible, and the lab is that argument built at the scale of one human and one machine. Agency sits at the center of this flywheel, and the flywheel is designed to strengthen its center with every turn. Each cycle returns capacity in both directions, sharper scenarios for the system and restocked judgment for me, which is the precise opposite of how psychological debt accrues.
In Becoming a Creator of Creators I gave this orientation its name and its test: after each cycle of work, are the people involved more capable of creating, or less? The lab forces me to apply the test to myself, because I am the only person in some of those loops. After each cycle, is the judgment in the room sharper or duller? Psychological debt is the precise inverse of the test passing.
The honest score so far is mixed in an instructive way. Some sessions I catch Ana converging, three scenarios wearing the same assumption in different clothes. Some sessions I discover I was the one converging, nodding along with fluent output I should have pushed against. Both entries go in the ledger, because a record of only the first kind would be marketing.
One flag in fairness to the research: Champniss runs a behavioral science consultancy, so a framing where the remedy is behavioral intervention also describes his business, the same way my framing that the remedy is architecture describes mine. The data sits apart from both frames, and what it plainly says is that the way most enterprises deploy AI is depleting their people invisibly.
Two ledgers, and a third
The enterprises that finish AI rather than merely starting it will keep two ledgers, the financial one they already keep and the human one this research shows them how to read. I am keeping a third, the personal one.
When I introduced the Regenerative Prosperity Model in Regeneration as Operating System, I put Self first among its five domains, ahead of Nature, Community, Design, and Capital, because the builder is part of the system and the condition of the builder shapes the condition of the company. The third ledger is that first domain receiving its first honest bookkeeping. I would rather record my own depletion early than discover it the way the soil discovered industrial agriculture, in the season the ground gave out.
I built a career climbing rungs that today would be automated before I reached them. The people climbing after me deserve an architecture that puts the rungs back. What I had not planned to write, until that thirty-minute build, is that the people who already climbed need the rungs too. The ladder, it turns out, was never something you finish.
The field remains open. This is the first thing that walked into it, and it walked straight up to me. The protocols are young, Ana will have new scenarios for me by morning, and the friction they meet will be mine. I will return to this page when the evidence does.
---
Champniss, G. (2026). The Psychological Costs of Adopting AI. Harvard Business Review, May 1, 2026. Related work referenced above: Institutional Evolution in the Age of Automated Execution (451°, January 2026), and earlier essays in this journal, including Soil, Silicon, and the First Architecture of Wonder, Building the Company, The Wedge: Autonomous Risk Intelligence, Trust Is Integrated, Not Declared, Regeneration as Operating System, Becoming a Creator of Creators, and The Field Remains Open.