2026-05-23
May 23, 2026
Issue 017 — The half-life of a technical skill has collapsed to under five years. The only career capital that compounds in 2026 is the part of you no model can run: judgment, taste, and the ability to decide what "good" means.
There's a particular anxiety I keep meeting in senior people this year. A staff TPM who has rebuilt her entire workflow around three agentic tools since January, and is quietly terrified that a fourth will obsolete the second one by Q3. A director who got his team fluent in one orchestration framework right as the field standardized on a different one. A VP who can feel that the thing he was known for — running the tightest cross-team status machine in the org — is now something a model does in the background while he's still scheduling the meeting.
The instinct in all three cases is the same: learn faster. Add the new tool, take the cert, rebuild the workflow again. It feels like progress because it's effort, and effort against the newest thing feels like staying current. But it's the wrong move, and the reason is a number most senior leaders haven't internalized: the half-life of a specific technical skill has fallen from the 10–15 years it held a generation ago to under five today, and for AI-tooling skills it's measured in months. The World Economic Forum's Future of Jobs Report 2025 puts hard edges on it — employers expect 39% of workers' core skills to change by 2030.
If your skills are depreciating that fast, the answer is not to run the treadmill faster. It's to stop measuring your worth by the things on the treadmill — and to know which of your skills appreciate instead.
Last Saturday this newsletter argued that sponsorship — the relational capital that gets you pulled into rooms — is engineered, not given. Today's argument is about the other half of a career: the skill capital you bring once you're in the room. And the central fact of 2026 is that not all skill capital is the same kind of asset. Some of it compounds. Most of what we train hardest on depreciates.
Start with the mechanism, because it explains everything downstream. When the cost of producing something collapses, value migrates from production to evaluation. AI has made the production of code, copy, decks, analyses, and plans nearly free. So the scarce, valuable, paid-for act is no longer making the artifact — it's deciding which artifact is worth making, judging whether the output is good enough, and sensing when the whole direction is wrong. Andrej Karpathy put the sharpest version of this at Sequoia's AI Ascent earlier this year: you can outsource thinking, but you cannot outsource understanding. The same talk reframed taste not as an aesthetic luxury but as your internal evaluation function — the quality of the judgment you run every time you compare what's in front of you against a standard and decide good enough or not yet. When a machine can generate ten options in a second, the person who can correctly pick the right one is the entire job.
Now hold your own skills up against that and sort them into two piles.
The depreciating pile is everything tied to a specific tool, surface, or rote process. Fluency in one model's API. A particular prompting style. Deep expertise in a single PM or roadmapping tool. And — uncomfortably for our discipline — the classic operational core of program management: manual status aggregation, dependency-tracking by spreadsheet, meeting choreography, the weekly report assembled by hand. These were genuine skills. They are also exactly the tasks AI absorbs first. Investing your scarce learning hours here buys an asset that's worth less every quarter.
The compounding pile is everything that gets more valuable as production gets cheaper: judgment under ambiguity, taste and evaluation, systems and program thinking, written reasoning, stakeholder navigation, and the ability to define "good" precisely enough that others — humans or agents — can execute against it. The WEF data lines up almost eerily with this: the core skills employers rate as most durable through 2030 are analytical thinking, resilience and adaptability, and leadership and social influence — not one of which is a tool. The fastest-growing skills (AI and big data, tech literacy) matter too, but read them correctly: they're a layer you must continually refresh, not a moat you build once. Treat AI fluency as maintenance, not strategy.
For senior TPMs and tech leaders the tell is even cleaner. Tanya Reilly's The Staff Engineer's Path defines staff-level impact as three things: big-picture thinking, executing work that crosses many teams, and leveling up the people around you. Every one of those sits squarely in the compounding pile. None of them is something you install. That is not a coincidence — it's why the senior IC and leadership ladders survive each tooling cycle while the job titles underneath them churn.
So the move is not "learn less." It's to rebalance the portfolio. Most people, under treadmill anxiety, spend the overwhelming majority of their learning time on the depreciating layer because it's concrete and gives a fast hit of competence. The compounding layer feels slower and vaguer, so it gets the scraps. Flip the ratio. Spend the bulk of deliberate effort on judgment, taste, and written reasoning — and the good news is these are trainable, not innate. Taste improves the same way any evaluation function does: reps. Judge AI output against an explicit standard, predict which option is best before you check, write down why, and notice when you were wrong. That loop, run weekly, compounds into the one skill no model will rent you.
Try this week. Put your top ten professional skills on a single page. Mark each C (compounds — more valuable as production gets cheaper) or D (depreciates — tied to a specific tool or rote task AI is absorbing). Then estimate the share of your last quarter's learning time that went to each. If most of it went to the D's, you're running the treadmill. Pick one C — judgment, taste, written reasoning — and give it the larger share next quarter.
What it is. The argument that a great career is bought, not found: you accumulate career capital — rare and valuable skills — and then spend that capital to shape your work into something you control. Its companion is the craftsman mindset: focus on what you can offer (and on getting relentlessly better at it through deliberate practice) rather than on what the job offers you. Newport's frame is the cleanest lens for deciding which skills are worth your finite learning hours.
When to use it. When you feel the treadmill anxiety — chasing every new tool and still feeling behind. At promo planning, a role change, or any moment you're deciding what to invest the next two quarters of learning in. Especially useful when a skill you were known for is being commoditized and you need to choose what to become known for next.
How to run it (applied as a quarterly skill audit):
When NOT to use it. Don't use it to justify ignoring AI tooling entirely — the maintenance layer is real, and falling off it has its own cost. And don't apply the craftsman mindset to a domain you're leaving; capital is domain-specific and doesn't transfer cleanly across a career pivot.
Example: a staff TPM known for the org's tightest status-reporting machine sees that capital depreciating as agents automate the reporting. She redirects her quarterly learning from a fourth tracking tool toward written decision memos and cross-team architecture judgment — and gets pulled onto the platform strategy she'd been locked out of within two quarters.
The Future of Jobs Report 2025 — Skills Outlook — The World Economic Forum's data on the 39% core-skill churn by 2030, and the ranked list of what stays durable: analytical thinking, resilience, leadership and social influence. The reference table to bring to your next career or team-development conversation.
Where Is Technical Program Management Heading in 2026? — A clear-eyed read on how the TPM role is bifurcating: the coordination layer thins as AI absorbs it, while the strategic-judgment layer is where the senior roles concentrate. The depreciating-vs-compounding split, applied directly to our discipline.
Sequoia AI Ascent 2026: Andrej Karpathy — The "outsource thinking, not understanding" talk, and the case that taste — your internal evaluation function — is now the bottleneck skill. The most useful articulation this year of why judgment is the asset that compounds.
"In times of change, learners inherit the earth, while the learned find themselves beautifully equipped to deal with a world that no longer exists."
— Eric Hoffer, Reflections on the Human Condition (1973)
Don't miss what's next. Subscribe to Critical Path: