← All posts Essay

The Talent Famine: How AI Is Eating Its Own Future Workforce

· 12 min read

The headlines of 2026 tell a tidy story: AI makes software engineers more productive, so companies need fewer of them. Layoffs at Meta, Microsoft, Amazon, Atlassian, Oracle, GM. Tens of billions saved. Stock prices up. Roadmaps shipping faster than ever.

It's a clean narrative, and it's almost entirely about the next four quarters.

What it ignores is something that won't show up on a CFO's spreadsheet until 2030 or later: the industry is dismantling the very ladder it uses to produce the senior engineers it now depends on. The same companies racing to replace junior developers with Claude, Codex, and Cursor are quietly burning the only mechanism by which seniors get made. We are in the early months of a famine being engineered in plain sight — and when the bill comes due, very few of today's decision-makers will still be in their seats to pay it.

1. The cuts are real, and they aren't slowing down

As of May 2026, more than 92,000 tech workers had been laid off year-to-date, on top of roughly 800,000 since 2020. April 2026 alone was the worst single month for tech cuts in two years, with over 45,000 positions eliminated in thirty days. Tech is now the only major industry where layoffs are accelerating year over year — every other sector is cutting less than in 2025.

Crucially, this is not the post-ZIRP overhiring correction we kept being told it was. Meta, Microsoft, Alphabet, and Amazon are projected to spend roughly $700 billion combined on AI infrastructure in 2026 — while simultaneously announcing the most aggressive workforce reductions in their histories. Meta's AI budget alone is reportedly four to five times its total annual payroll bill. These companies are not struggling. They are explicitly choosing a future with fewer humans and more silicon.

The pattern is no longer concentrated in customer support, recruiting, or "non-core" functions. GM cut 500–600 IT workers in May 2026 and openly said it would rehire for "AI-native development, data engineering, cloud engineering, model and agent design, and prompt engineering." Atlassian cut 1,600 people — about 10% of its workforce — five months after its CEO had publicly promised to hire more engineers over the next five years. The share of layoff announcements explicitly tied to AI restructuring has tripled since spring 2025.

This is not a cycle. This is a phase change.

2. The vanishing ladder

The dirty secret of those layoffs is that the cuts are not evenly distributed across seniority. They almost never are.

Stanford's Digital Economy Lab, working with ADP payroll data, found that employment for software developers aged 22 to 25 has fallen nearly 20% from its late-2022 peak, while employment for developers over 30 has held steady or grown. A separate Harvard study tracking 62 million workers across 285,000 firms found that when companies adopt generative AI, junior employment drops 9–10% within six quarters. Senior employment barely moves.

The compensation platform Ravio reports that entry-level hiring rates (P1 and P2 bands) dropped 73% year-over-year as AI automates routine work. Industry analyses put the junior share of new tech hires at roughly 7%, down from 15% three years ago — meaning for every 100 new technical employees, where companies once hired 15 juniors, they now hire 7. CS graduate unemployment sits at 6.1% versus 3.6% for the general workforce. The dev.to community has coined a brutal phrase for the situation: "We're the first generation that has to be Senior before we're allowed to be Junior."

The downstream signal is already visible. Forrester forecasts a 20% drop in computer science enrolments in 2026 as prospective students respond to the deteriorating market. Internship postings in tech are down 30% since 2023. The people who would have been the senior engineers of 2032 are right now choosing finance, biotech, healthcare, and law instead — fields that still have visible ladders with bottom rungs intact.

And the mid-level engineers — the ones who survived the 2022–2024 cull — are being compressed too. Internal Klarna-style experiments and AI-tool rollouts mean teams that needed ten engineers now run with four, and the ones cut from ten to four are rarely the most senior. Many mid-level engineers, watching this happen around them, are simply walking away from the career while they still can.

3. The new rockstars: senior engineers as one-person product teams

While the bottom of the pyramid is being demolished, the top is being supercharged.

A senior engineer fluent in Claude Code, Codex, Cursor, and the rapidly maturing agent ecosystem now ships in a sprint what a senior plus a junior used to ship in two. In a growing number of startups, that senior is the engineering team. Solo founders backed by frontier coding models routinely report week-over-week growth rates of 10% — five times the rate of the classic two-percent-per-week high-growth benchmark.

The market is pricing this rather precisely. AI-focused senior engineers now command base salaries of $200K–$310K, with total compensation at frontier labs running $300K–$490K for seniors and $500K–$750K+ for staff. OpenAI is reportedly paying $300K retention bonuses to new-grad technical hires. Meta has offered sign-on packages exceeding $100 million for elite AI researchers. MRJ Recruitment's 2026 benchmarks show 9% year-over-year salary growth for engineers with three-to-five years of hands-on ML experience — the steepest climb of any experience band, because these are the people who can actually ship production systems without supervision and don't yet cost what a staff engineer costs.

This is what the market is signaling, loudly: people who can use AI well, on top of real engineering experience, are now multipliers. Everyone else is a cost line being optimized against.

4. Why seniors still matter (and why AI doesn't replace them)

Here is where the executive napkin-math breaks down.

AI is genuinely good at the surface layer of software: boilerplate, scaffolding, predictable refactors, test generation, line-by-line translation of well-specified tickets into pull requests. AI completes routine code roughly 55% faster than humans. But boilerplate was never the bottleneck. The bottleneck is, and remains:

Independent measurements from a 39,000-developer study suggest the actual productivity uplift from current AI coding tools is around 2.1%, with software delivery performance declining 7.2% in some studies — a fraction of the 24–25% gains many executives believe they're getting. Code churn (code rewritten or deleted within two weeks) has roughly doubled. Duplicate-code rates are up four-fold. Up to 30% of AI-generated snippets ship with security issues. The model can write the code; it cannot yet own the consequences of the code.

That ownership — the part that actually decides whether software works — still requires an experienced human. The painful irony is that this is precisely the human the current hiring strategy is making sure we will not produce.

5. The pipeline is being unbuilt

This is the part the quarterly earnings call doesn't capture.

The path from a CS graduate to a competent senior engineer takes seven to ten years. Those years are not abstract. They are made of code reviews from someone slightly more experienced, of architecture decisions you watched a staff engineer make and questioned afterwards, of the on-call shift where you broke production at 3am and learned what "idempotent" really means. None of that happens to a prompt. It happens to a person, embedded in a team that is willing to spend time on them.

We are now in the third consecutive year of companies declining to make that investment. Salesforce's CEO publicly stated the company would hire essentially no new engineers in 2025. Two-thirds of global enterprises in a 2025 IDC/Deel survey said they would slow entry-level hiring; 91% said roles were already changing or disappearing because of AI. Seven in ten survey respondents reported fewer on-the-job development opportunities for junior employees. Universities are producing fewer CS graduates. Bootcamps are pivoting away from coding toward "AI prompting." The intake to the pipeline is shrinking, the apprenticeship layer of the pipeline is being demolished, and the mentorship that turns junior people into senior people is being replaced with the assumption that a senior plus a model is enough.

For a year or two, it might be. For five to ten years, it cannot be.

6. The demographic time bomb

Look at the age distribution of the senior engineers currently propping up most companies' codebases. In most organizations the picture is something like this:

A material share of today's senior individual contributors and architects entered the field during the 2000s build-out. They are now ten to fifteen years from retirement. Some will stay longer, some will leave earlier. But the mass of deep, hard-won, production-tested engineering judgment in most companies is concentrated among people who are approaching the end of their working lives.

There is no comparably-sized cohort behind them. There won't be, because the cohort that should be behind them was told in 2024–2026 that the industry no longer had room for them.

By roughly 2029–2031, you can expect the math to land all at once: a wave of retirements colliding with a missing decade of mid-level engineers, in a labor market where demand for sophisticated, AI-orchestrating senior engineering only keeps growing. One projection — perhaps optimistic — already puts the global software engineering shortfall at 85 million people and $8.4 trillion in foregone revenue by 2030.

7. The price shock no one is pricing in

When demand outgrows supply in a market with no near-term substitute, the only adjustment variable is price.

We are already seeing the leading edge of this: 9% annual base-salary growth for the most in-demand mid-senior band, 70%+ AI pay premiums at the senior level inside specific firms, $100M sign-on offers at the frontier, 114-day time-to-fill for senior AI engineering roles versus 52 days for the broader tech market. These are the numbers in a market where the supply of senior engineers is still mostly intact, because the people earning those salaries were trained before the apprenticeship ladder collapsed.

Now extrapolate. By the early 2030s, an "AI-native senior engineer who can run a product end-to-end" will be the most strategically important role in most companies — and the supply will be smaller, older, and shrinking. Once the market actually understands the constraint, expect compensation for that profile to rise not 9% a year but multiples. 2× to 5× of today's rates is not a wild estimate; it is what every other labor market in history has done under comparable supply collapse. Petroleum engineers in the 1970s, COBOL maintainers in the late 1990s, cybersecurity specialists in the mid-2010s — all the same story, all triggered by the same structural mismatch we are now actively creating.

The companies that decided in 2025 that they didn't need to train juniors will, in 2032, be the same companies bidding against each other for the seniors they refused to grow.

8. The endgame: AI labs as the new software outsourcers

This is the part of the picture most discussions miss, and it is the part that matters most strategically.

If the global supply of senior engineers contracts while AI-built software demand explodes, the bottleneck does not just become a hiring problem — it becomes a services problem. Companies that cannot hire the talent will buy it. And the entities best positioned to sell it are the AI labs themselves.

This is not speculation. It is already the announced strategy.

In May 2026, OpenAI launched the OpenAI Deployment Company ("DeployCo"), a majority-owned subsidiary backed by more than $4 billion in initial investment. Its product is not a model. Its product is embedding engineers directly inside enterprise customers to build production AI systems around their workflows. To staff it, OpenAI acquired Tomoro, bringing roughly 150 Forward Deployed Engineers and Deployment Specialists onboard from day one. Engagements are reported to start at $10 million per client. Anthropic runs an analogous "Applied AI Engineering" function. Cohere, Databricks, and Scale AI all do versions of the same thing. Job postings for "Forward Deployed Engineer" have spiked roughly 800% in the past two years.

The model is borrowed almost verbatim from Palantir, and its strategic genius is the lock-in. When a frontier lab's engineering team spends six months building a custom AI system that is deeply wired into a customer's data, workflows, compliance architecture, and operational metrics, that system becomes load-bearing infrastructure. You don't rip it out. You depend on the lab that built it — for maintenance, for upgrades, for next year's model, forever. It is far stickier than any SaaS contract.

Combine these two trends — a shrinking pool of senior engineers and AI labs aggressively positioning as the world's premium software-development-as-a-service vendor — and the destination becomes uncomfortable to look at directly. Custom enterprise software, the largest single category of global IT spending, is on a path toward being effectively intermediated by two or three AI labs. Not because they will write a single line of code you see, but because they will own the engineers who do, the models they use, and the integration layer between your business and both.

That is not a productivity story. That is a structural redistribution of where software economic value is captured.

What this adds up to

The 2026 conversation about AI and engineering jobs is being framed as a productivity revolution. It is also a market-structure revolution that almost nobody on the demand side is pricing in.

The companies cutting juniors today are optimizing for a 2026 P&L and accepting, mostly without realizing it, three liabilities that come due later:

  1. A senior engineer shortage they helped create by closing the entry door
  2. A compensation explosion for the seniors they will still need
  3. A growing dependence on AI labs as the de facto outsource layer for the software work they can no longer staff internally

The painful part is that each of these is rational at the individual firm level. No single CFO is wrong to cut three juniors and a copilot license against one senior plus AI tooling. The damage shows up only at the aggregate, civilizational level — and by the time it is legible in the data, it will be too late to fix on the timescales required.

There is a counter-strategy. A handful of long-lived enterprise software companies, infrastructure firms, and a few self-aware large employers are already quietly increasing junior hiring in 2026, on the explicit theory that future seniors must come from somewhere. They will look like idiots for two or three more years. Around 2030, they will look like the only adults in the room.

The talent famine is not coming. It has already started. We are simply still in the part of the curve where the cuts feel like efficiency.

Idea by Max Kudinov. Researched and written with Claude.

Sources: CNBC; Stanford Digital Economy Lab (2025–26); Ravio Compensation Trends 2026; Forrester 2026 Predictions; MRJ Recruitment 2026 benchmarks; SignalFire; Harvard "Canaries in the Coal Mine" study; IDC/Deel 2025 enterprise survey; OpenAI corporate communications (May 2026); MindStudio analysis on FDE strategy; Layoffs.fyi tracker; Challenger Gray & Christmas; IEEE Spectrum; Stack Overflow Developer Survey 2025; Glassdoor; Levels.fyi.