Your skills are assets, and every asset has a depreciation schedule. Some of what you know is losing value on a fast clock — a specific tool's API, a platform's current quirks, the framework everyone hired for two years ago — some is barely decaying at all, and a small, precious category is actually appreciating as the environment shifts. Most people never run this accounting. They manage a career the way an amateur holds a stock: buy once, feel productive for having bought, and never check the position again while it quietly rots to zero. The discipline that fixes this is boring and borrowed wholesale from finance. Treat your skills as a portfolio, know the depreciation rate on each holding, and rebalance on purpose.
I want to be precise about the mechanism, because "depreciation" is doing real work here, not decorating a metaphor. A depreciating asset loses value on a schedule tied to something external. Accounting models it two ways. Straight-line: the asset sheds a fixed amount of value each year until it hits zero. Declining-balance: it loses a fixed percentage of its remaining value each period, which is exponential decay — most of the loss front-loaded, a long thin tail. Human skills are almost always the second kind. A hot framework doesn't lose 20% of its value every year for five clean years; it loses most of its market value in the first couple of years after something better ships, then lingers as legacy maintenance work at a fraction of its peak price. That's a half-life, and it's the single most useful number you can attach to anything you know.
The half-life is the whole game
Put a half-life on a skill and its true economics stop hiding. Say a particular tool's fluency has a half-life of two years — a fair estimate for anything governed by a vendor's release cadence. Learn it to mastery today and, holding your effort constant, you retain roughly half its market value in two years, a quarter in four, an eighth in six. Six years out you're carrying 12% of what you paid for, still listing it on your resume as if it were the full asset.
Meanwhile the reason a false-discovery-rate correction exists — what it actually protects you from, when a batch effect is silently confounding your signal — has not lost measurable value in the fifteen years since I first needed it. I've watched the specific bioinformatics stacks I learned churn completely: aligner versions, the single-cell package whose entire API got rewritten between major versions, the differential-expression idioms that were table stakes and are now trivia. The statistical reasoning underneath all of it never depreciated at all. Same hours invested, wildly different schedules. Nobody told me at the time which was which, and that's the expensive part.
Here's the trap, and it's structural, not a matter of laziness. Fast-depreciating skills are the ones that feel like progress. They're legible — you can name the tool, list it, get certified in it. They're liquid — they're literally what job postings screen for, so they convert to offers now. And they're concrete — you can feel yourself getting better at them hour by hour. The slow-depreciating and appreciating skills have none of those properties. Judgment, taste, the ability to specify a problem so precisely that someone else could execute it, systems thinking, raw learning velocity — the payoff is diffuse, delayed, and unattributable. You can't point to the Tuesday you got better at judgment. So a rational person, optimizing for signals they can see, loads the portfolio with fast-decay tactical skills and starves the compounders. It's hyperbolic discounting wearing a career-development costume: the immediate, legible reward beats the larger, later, fuzzier one every single time you're not paying deliberate attention.
Three buckets
You don't need a spreadsheet with decay constants. You need to sort what you know into three buckets by depreciation rate, and the sorting alone will show you how lopsided your position is.
| Bucket | What's in it | Half-life | Why you hold it |
|---|---|---|---|
| Fast-decay tactical | A tool's API, a framework's idioms, a platform's current quirks, a specific vendor's UI, this quarter's hot library | ~1–3 years | Cash flow. It gets you hired and opens doors now |
| Slow-decay durable | Judgment under ambiguity, taste, clear writing, systems thinking, statistical reasoning, selling, negotiation | ~10–20+ years | The core of a career; compounds quietly |
| Appreciating meta | Problem specification, learning velocity, knowing what's worth building, reading a system for its failure modes | Rising | Gets more valuable as tools change and execution automates |
The third bucket is the one worth staring at, because "appreciating" is a strong claim and I mean it literally. An appreciating skill is one whose market value rises as the environment changes — the environment does the compounding for you. Problem specification is the cleanest example. As models and agents drive the cost of execution toward zero, the scarce input becomes a precise statement of what you actually want and how you'd recognize it when you saw it — which is exactly why prompt engineering depreciates while problem specification compounds. The better the engine, the more the quality of your destination matters. That's not a skill decaying slowly; it's a skill whose worth is a positive function of the very trend killing your tactical skills. The same force does both.
Automation is a repricing event, not a slow tide
This is where the portfolio frame earns its keep, because we're living through a fast, uneven repricing of exactly one bucket. Automation isn't depreciating all skills evenly. It's slamming the execution tier specifically. Once you see that a job is a bundle of heterogeneous tasks and that agents dissolve jobs into tasks rather than replacing whole roles, the mechanism is obvious: the automatable tasks are precisely the codifiable, executional ones — the drafting, the boilerplate, the first-pass analysis — and those are the tasks your fast-decay tactical skills were priced on. So the half-life on the entire tactical bucket just got shorter. A skill you'd penciled in at a three-year half-life might now be running on eighteen months, not because you got worse at it but because the thing that made it scarce got cheap.
The correct response to a repricing is not to panic-sell the whole position. It's to rebalance toward the assets whose value the same event is raising. When the cost of execution collapses, the return on judgment, taste, and specification goes up, because now you can direct near-free execution and the only bottleneck left is knowing what's worth executing and whether the output is right. The distribution gets brutal in a specific direction — and here I'm forecasting, not reporting: people whose value was fluent execution get compressed toward the machine's baseline, and people whose value was specification and judgment get amplified by the exact same technology. Which bucket you're overweight in decides which side of that line you land on.
What to actually do
Run the audit, then reallocate, then ladder. Three moves.
Audit by depreciation rate, not by pride. List the skills you actually used in the last month — the honest list, the things you did, not your job title. For each, ask the only question that matters: what has to change in the world for this to lose half its value, and how fast does that change ship? A vendor's release cycle governs a tool skill. A framework's dominance governs its idioms. Nothing external governs your judgment except your own decline. Sort into the three buckets. Almost everyone discovers they're 80% weighted in bucket one and telling themselves it's a diversified portfolio.
Fund the slow and appreciating buckets even though the payoff is delayed. This is the whole discipline, and it's hard for the same reason saving is hard: you're moving resources from a visible present reward to an invisible future one. Concretely, carve out a fixed fraction of your learning hours — at least a third — and spend it on things that produce no certificate this quarter. Write something for a hostile reader. Make a real decision, write down your reasoning, and go back in six months to check whether you were right; that loop is how judgment is actually trained, and nobody does it because nothing forces them to. Practice specifying a problem tightly enough to hand off. These lose every scheduling fight to the tactical skill with the immediate payoff, so they have to be scheduled, not left to willpower.
Ladder your maturities so your skills don't all expire in the same market shift. Bond investors don't put everything into one maturity date, because then a single rate move at that date wrecks them. They build a ladder — staggered maturities — so they're never fully exposed to one moment. Your skills need the same structure, and the failure mode is correlation. Five skills all inside one tool's ecosystem aren't five bets; they're one bet, and they all crash together the day that ecosystem loses favor. I've watched engineers whose entire portfolio was correlated to a single frontend framework get repriced overnight when the industry moved, and they mistook their five expiring skills for a diversified career. Hold tactical skills across uncorrelated domains, and anchor the ladder with durable and appreciating skills that no single shift can call due at once.
Keep the fast-decay skills. That's not a concession, it's the plan — they pay the bills, they're the price of being credible and employable, and you cannot get hired on "good judgment" with nothing shipped in a current stack. Just hold them consciously, as a consumable you will rebuy every couple of years, and stop mistaking the act of accumulating them for the act of building something that lasts. The amateur buys the stock and never looks again. The professional knows what he owns, what it's worth, and how fast it's bleeding. Go look at your position.