Your Knowledge Is a Depreciating Asset. Judgment Compounds.

For a decade we got paid for what we knew. That asset is being written down to zero. Here is the one that is appreciating instead, and how to move your weight onto it before the market finishes repricing it.

For about a decade, the reliable way to build standing in technical work was to accumulate reproducible knowledge and rent it out. You learned the tool, the syntax, the runbook, the migration path, and you got paid for knowing it, whether as a salary, a course, a book, or a paid community built around a library of how-to. The knowledge was the inventory. It had value because it was scarce and you held more of it than the person paying you.

That inventory is being written down to zero. Not because anyone mismanaged it. Because the thing on the shelf, reproducible knowledge delivered on demand, is now generated for free in seconds, personalized to the exact situation, by a tutor that never tires and updates the moment the tool does. An engineer who needs a working pipeline no longer buys the lesson. They describe the constraint and get a tailored answer for less than the cost of looking it up.

This is an asset problem before it is anything else. If your balance sheet, career or business, is concentrated in reproducible knowledge, you are long a deflating asset. The move is not to accumulate faster. It is to work out which asset is appreciating and shift your weight onto it before the writedown is fully priced in. That asset is judgment. Here is the line between the two, drawn with the data.

The threat, named#

Call it Knowledge Deflation: the rapid collapse in the market value of reproducible technical knowledge as AI makes its transfer free, instant, and personalized.

Technical knowledge always decayed. Docker made the VM courses obsolete. Kubernetes turned the Swarm tutorials into a dead end. Every few years a new layer arrived and last year’s training material aged out. Producers stayed ahead by shipping the next thing before the last one fully rotted. AI breaks that arithmetic. It does not swap one body of material for another. It shortens the useful life of all of it at once, faster than anyone can produce the replacement, and it does so on the demand side too. The buyer who would have paid for the explanation now generates a bespoke version of it for the price of a prompt.

The deflation is already showing up where knowledge gets sold directly.

Copy/pasted lines rose from 8.3% of changed code in 2021 to 12.3% in 2024, while refactored (“moved”) lines fell from about 25% to under 10% over the same window. 2024 was the first year copy/paste outpaced refactoring in the dataset.

— GitClear, AI Copilot Code Quality 2025 (211 million changed lines, 2020–2024)

Read that as an economic signal, not only a quality one. When knowledge is cheap to reproduce, it gets reproduced instead of reused. The same force that fills a codebase with near-duplicate blocks fills the market with near-duplicate explanations. Abundance is exactly what destroys the price.

The blast radius#

The deflation runs through three balance sheets at once: the educator’s, the engineer’s, and the trust of the person doing the buying.

On the supply side, the people who sell learning for a living are watching demand erode in real time. I watched the founder of an interactive-coding-course platform publish his own keyword research while deciding what to build next, and the read was sobering for anyone in that business. Search demand for “learn to code” queries is draining as the audience routes its questions to an AI assistant instead of a curriculum. When the person whose income depends on selling courses tells you the top of the funnel is emptying, that is not doom-posting. That is a supplier reading his own order book.

The macro picture agrees. US computer science enrollment fell 8.1% in 2025 after more than a decade of growth, the steepest one-year drop of any major in six years. CS slid from the fourth to the sixth most popular field of study. It is the first cycle where the decline is pinned on AI reshaping the work itself, not on a dot-com-style boom and bust. The pipeline that fed the entire knowledge-transfer economy is narrowing because the buyers can sense the inventory is depreciating.

And the asset is not only deflating. It is unreliable at the margin, which speeds the writedown.

Trust that AI output is accurate fell from 40% to 29% year over year, the lowest the survey has recorded. Favorability slid from 72% to 60%. The top frustration, named by 45% of developers, is “AI solutions that are almost right, but not quite,” and two-thirds now say they lose time cleaning up that almost-right code.

— Stack Overflow Developer Survey 2025

“Almost right, but not quite” is the tell. Commoditized knowledge is now cheap and untrustworthy at the margin, which means the value does not vanish. It relocates. It moves to whoever can tell looks correct from is correct. That is not a knowledge skill. It is judgment.

The most direct evidence that the machine cannot hold the high-value end is what happens when you delegate real work to it.

Across 19 models and 52 professional domains, even frontier systems (Gemini 3.1 Pro, Claude 4.6 Opus, GPT 5.4) corrupted an average of 25% of document content over long, multi-step workflows. Agentic tool use did not help. Degradation got worse as documents grew larger and interactions ran longer.

— Microsoft Research, “LLMs Corrupt Your Documents When You Delegate” (DELEGATE-52), arXiv, April 17, 2026

Transfer of knowledge: free, instant, and increasingly good. Sustained application of judgment to a real, evolving artifact: still failing a quarter of the time at the frontier. That gap is the whole investment thesis.

Why this is a balance-sheet question, not a skills question#

The instinct, when knowledge deflates, is to learn faster and accumulate more inventory more aggressively. That is the wrong move, and the data above says why. You would be doubling down on the asset that is losing value, in a race against a free competitor that restocks instantly. Speed of accumulation is not the lever.

The lever is what you choose to be paid for.

Reproducible knowledge answers “how does X work” and “how do I configure Y.” The market price of that answer is now close to zero, because the answer is generated on demand. Applied judgment answers a different question: given our compliance posture, our existing sprawl, the two teams that own overlapping systems, and the fact that we cannot take downtime this quarter, what should we actually do, in what order, and what breaks if we are wrong? An AI assists with that. It cannot own it. Owning it requires context the model does not have, accountability it cannot hold, and the scars to know which of three plausible answers survives contact with production.

That is why judgment is the appreciating asset. Every junior task the machine absorbs raises the premium on the senior judgment it cannot. I made the organizational version of this argument: teams that thin out senior review to chase velocity lose the people who can tell a working system from a plausible one. The cheaper reproducible knowledge gets, the more valuable it becomes to deploy that knowledge correctly under real constraints. The two assets are moving in opposite directions, and most people are still weighted into the one going down.

The same writedown hits teaching twice#

If you sell knowledge as a product, courses, tutorials, library-style communities, the deflation hits you from two directions at once.

From above: AI commoditizes the core inventory. The recorded explanation competes with the live, free, personalized one, and loses on price. Producing more modules is running faster on a treadmill that keeps accelerating. The tell is in the product itself. When the navigation of a learning offering fills with “planned,” “building,” and “coming next,” it has stopped being a library and become a promise of future content, sold today. That works while content is scarce. It stops working when content is abundant and free.

From below: imitation. I have paid into a handful of these over the past couple of years, and a second pattern repeats alongside the first. Someone with real authority builds something that works, and a wave of imitators ships a thinner version of the same thing. The copies compete on the one axis AI has already won, reproducible knowledge, without the audience, the production scars, or any reason to exist that a model cannot satisfy for free. The category gets squeezed from both ends. AI deflates the value from the top, imitators dilute what is left from the bottom. The strongest operators feel the ceiling. Everyone copying them hits it sooner.

The defensive move is not to produce faster. It is to stop selling the answer and start selling the room where answers get worked out. The durable version of a technical community in this era is not a content library with a forum bolted on. It is a standing conversation among people with enough production experience to separate signal from noise, organized around the questions that do not have a recorded answer yet. Not “how do I configure this,” but “what survives when the thing I configured can be generated by anyone in thirty seconds.” The members are the asset. The host curates the question and holds the standard. That is harder to sell and much harder to fake, which is why it holds value while the library does not.

Control protocol#

How to move your weight off the deflating asset and onto the appreciating one. Career and business, same five moves.

1. Stop pricing reproducible knowledge. Price judgment. Audit what you currently get paid for. Every line item that is “I know how to do X” is exposed. Every line item that is “I decide what is worth doing, in your context, and I own the outcome” is defensible. Shift the mix deliberately.

2. Convert content from product to proof. A blog, a talk, an open-source project should no longer be the thing you sell. It is evidence of judgment that makes you more valuable at the thing you do sell, priced as a salary, then as consulting, then as advice. Give the knowledge away. It is worth nothing to hoard now anyway. Charge for the judgment it points to.

3. Move toward the work the machine fails at. DELEGATE-52 and the trust data point at the same high ground: sustained, accountable, context-heavy decisions over long horizons. Solutions architecture, security and compliance judgment, system design under real constraints. The hiring I watch is already concentrating there while execution-only roles deflate.

4. Make the supervisory layer the explicit job. If “almost right, but not quite” is the dominant failure mode, then catching it is now the highest-value work in the building. Whether you lead a team or sell your own time, fund and price that work as the deliverable it has become, not as a tax on velocity.

5. Stop competing on the commoditized axis. If your differentiation is “I have the knowledge,” you are competing with the free tutor and you lose on price. Differentiate on the axis AI cannot reach: ownership, accountability, and judgment earned in production over years.

Three questions to sit with this quarter#

Take them literally. The honest answers tell you how exposed your balance sheet is.

What share of my income today is paid for reproducible knowledge, the kind an AI assistant now generates on demand, versus applied judgment I am accountable for?
If every how-to I have ever produced became free and instant tomorrow, what would I still be uniquely paid for?
Is the content I am making a product I am trying to sell, or proof of judgment that makes me more valuable at what I sell? If it is the former, I am long a deflating asset.

Knowledge was the product. It is a commodity now, and the rational move is to plan your career and your business as if it is free, because functionally it is. The content era rewarded the people who could explain things clearly, and there was real money and status in being the clearest voice in the room. The AI era rewards what the content era undervalued: the ability to sit with a problem that has no recorded answer, weigh the tradeoffs, make the call, and own what happens next. You cannot download that. You cannot generate it from a prompt. You earn it in production, over years, by being accountable for outcomes.

Move your weight onto the asset that compounds. Judgment is the product now.

Vladimir Mikhalev

Docker Captain · IBM Champion · AWS Community Builder

The Verdict — production-tested analysis on YouTube.

YouTube GitHub LinkedIn