2026-05-24
May 24, 2026
Issue 018 — Coding agents took the grunt work that quietly turned juniors into seniors. The leaders who still have a staff-level bench in 2031 are the ones re-engineering the ladder now — deliberately, because it no longer builds itself.
The numbers that should keep senior leaders up at night aren't about model capability. They're about supply. SignalFire's State of Tech Talent Report 2025 found that the share of new graduates landing roles at the Magnificent Seven has dropped by more than half since 2022. A 2025 LeadDev survey found 54% of engineering leaders plan to hire fewer juniors because AI copilots let seniors cover more ground. New grads are stuck in the oldest trap in hiring — you need experience to get the job, you need the job to get experience — except this time the bottom rung has been sawn off rather than crowded.
Every recent issue of this newsletter has circled the same conclusion: in a world where production is nearly free, the scarce asset is judgment. Taste, evaluation, the ability to decide what "good" means. Everyone now agrees on this. What almost no one is saying out loud is the uncomfortable corollary: judgment is grown, not installed — and we are systematically automating away the exact reps that grow it.
AWS CEO Matt Garman said the quiet part loudly last year, calling the idea of replacing junior engineers with AI "the dumbest thing I've ever heard." His argument wasn't sentimental. It was a pipeline argument: "How's that going to work when ten years in the future you have no one that has learned anything?" That is the future-of-leadership question for 2026 — and it's a design problem, not a hiring problem.
For thirty years, becoming a senior engineer or a senior TPM was a byproduct. You got handed the unglamorous work — the migration nobody wanted, the flaky test suite, the status rollup, the integration grunt work — and somewhere in the friction of doing it badly, then less badly, you accumulated the pattern library that eventually reads as judgment. Niels Bohr's line is the whole mechanism in one sentence: an expert is a person who has made all the mistakes that can be made in a very narrow field. The mistakes were the curriculum. The grunt work was the classroom. Nobody had to design it; it fell out of the work.
AI removed the byproduct. Coding agents and copilots are spectacularly good at exactly the task-level, well-scoped, low-context work that used to be the junior's apprenticeship. The rational local move — and the data shows leaders are making it — is to route that work to an agent supervised by a senior, who is now 30–50% more productive, instead of to a junior who is slower and needs teaching. Each individual decision is defensible. The aggregate is a slow-motion capability collapse: an industry optimizing quarter by quarter for throughput while quietly defunding the only process that manufactures the senior judgment it has decided is priceless. Forrester projects a 20% drop in CS enrollment as students read the deteriorating signal. The funnel is narrowing at both ends at once.
This is the most consequential leadership tension of the decade, and it has a familiar shape: local efficiency versus systemic capability. It is the same trap as deferring maintenance, or cutting the training budget in a down quarter — each cut looks free because the bill arrives years later, on someone else's watch. The difference is that this bill is denominated in people who don't exist yet, which makes it almost impossible to see on any dashboard you currently run.
The reframe senior leaders need is this: apprenticeship has moved from a free byproduct of the work to a thing you must deliberately engineer. When the struggle that built deep knowledge is the first thing automated, you cannot wait for juniors to "pick it up by osmosis" — the osmotic surface is gone. You have to manufacture the experiences on purpose.
The most concrete answer on the table comes from inside Microsoft. Azure CTO Mark Russinovich and developer VP Scott Hanselman, in their 2026 paper Redefining the Software Engineering Profession for AI, name the problem precisely — AI is an accelerant for seniors and a drag on early-career engineers, so the economic gradient points straight at "hire seniors, stop hiring juniors" — and propose a counter-move they call preceptorship at scale. The model pairs early-career engineers with senior mentors in real product teams at ratios of roughly 3:1 to 5:1. Crucially, the pair uses AI together: the senior watches how the junior interacts with the agent — what they accept, what they reject, where their understanding breaks down — and the senior's role shifts from "the person who answers questions" to "the person who teaches judgment." The authors insist these run a year or longer, and — this is the part most orgs will skip — that mentorship be measured and compensated as a first-class deliverable, not volunteered in the cracks.
That last point is the whole leadership lesson. Under productivity pressure, anything not measured and funded does not happen. If "grew two engineers into independent operators" shows up nowhere in a senior's performance review or comp, it will lose every week to the shippable thing — and the bench will quietly empty. So the future-of-leadership job is not to resist AI in the name of juniors; it's to redesign the incentive system so that developing people is a funded output of the org, deliberately built back into work that no longer produces it on its own.
The orgs that figure this out get a compounding, durable advantage. By 2031 senior judgment will be the scarcest thing in the market, and you won't be able to buy your way out — everyone stopped making seniors at the same time, so the lateral market will be empty too. The only firms with a staff-level bench will be the ones that treated apprenticeship as infrastructure in 2026 and built it on purpose.
Try this week. Find one stream of well-scoped work you were about to hand entirely to an agent. Instead, route it through a junior-plus-senior pair who build it with the agent — the senior coaching judgment, not typing. Then check one thing: does "developing people" appear anywhere in your seniors' goals or comp? If it doesn't, you have already defunded your own pipeline, and that's the first fix — before any tooling decision.
What it is. A model for teaching the invisible parts of expertise — the reasoning, heuristics, and judgment that experts apply unconsciously and therefore rarely articulate. Born from the observation that traditional apprenticeship works because the work is visible, it adapts the master-and-apprentice loop to cognitive, knowledge work by deliberately making thinking visible. Its six methods are the practical blueprint for the "engineered apprenticeship" this issue argues leaders now have to build.
When to use it. Whenever the skill you need to transfer is tacit — architectural taste, "this estimate smells wrong," when to escalate, how to read a room — rather than a documented procedure. Especially valuable now that AI absorbs the procedural surface work that juniors used to learn from, leaving only the tacit, judgment-heavy core that this method is purpose-built to teach.
How to run it (six methods, in rough order):
When NOT to use it. For genuinely routine, well-documented procedures, this is overkill — write a runbook and hand it over (or hand it to the agent). Reserve the expensive human loop for the tacit judgment that can't be captured in docs.
Example: a staff TPM pairs a new hire on a messy cross-team dependency. Week one she narrates her own stakeholder calls (modeling); by week four the new hire runs them while she annotates afterward (coaching, articulation); by quarter's end he's surfacing dependencies she missed (exploration) — apprenticeship manufactured on purpose, from work AI couldn't have taught.
Microsoft's Russinovich and Hanselman Warn AI Is Hollowing Out the Junior Developer Pipeline — InfoQ's write-up of the preceptorship paper: AI as an accelerant for seniors and a drag on juniors, and the 3:1–5:1 mentorship model as the proposed fix. The clearest articulation of the problem and a concrete leadership response in one place.
The SignalFire State of Tech Talent Report — 2025 — The data behind the alarm: new-grad hiring at the Magnificent Seven down by more than half since 2022, demand for experienced engineers still rising. The chart to bring to your next headcount-planning meeting before you cut the junior reqs.
Do junior devs still have a path to senior roles in an AI age? — LeadDev on the practitioner reality of growing engineers when AI has eaten the on-ramp. The companion read for any leader who has to make the abstract pipeline argument concrete for their own team this quarter.
"An expert is a person who has made all the mistakes that can be made in a very narrow field."
— Niels Bohr
Don't miss what's next. Subscribe to Critical Path: