Something structural happened to professional-middle work in 2024–2026. Two squeezes at once.
From below, software engineers got more leveraged per head. One engineer is now doing what three did. Entry-level postings collapsed. Junior developer roles on Indeed fell roughly 40% from pre-2022 levels (Pragmatic Engineer). The share of Big Tech hires going to new grads went from 15% pre-pandemic to 7% in 2025 (SignalFire).
From above, middle management got flattened. Middle managers made up about 33% of all 2023 layoffs. Gartner projects that by end-2026, 20% of organisations will use AI to eliminate more than half their middle-management positions (Gartner).
Squeezed between these, a new job title: the AI engineer. Broader remit, ships across product, data, ops. At frontier labs, comp that can exceed $700K. LinkedIn added 75,000 AI-engineer postings in the US between 2023 and 2025 (WEF).
Here's what's actually happening.
The engineers
Indeed's software-developer postings index hit a five-year low in mid-2025: 36% below the February 2020 baseline (Indeed Hiring Lab). Entry-level tech hiring at the 15 biggest US tech firms dropped about 25% YoY in 2024 (SF Standard). CS graduates now have a 6.1% unemployment rate (NY Fed Q4 2025). CNN reported in August 2025 that CS grads averaged 150 applications per job search (CNN).
The individual-productivity story, honestly, is more complicated than the aggregate suggests.
GitHub's own Copilot study found a 55% task-completion speedup (2h41 → 1h11), with success rate improving from 70% to 78% (GitHub). METR's 2025 RCT on experienced developers using Cursor with Claude 3.5/3.7 on their own repositories found the opposite: 19% slower, with developers believing they had been 20% faster (METR). Both studies are real. They measure different things. The lift is big on greenfield and unfamiliar code. Negative on mature senior work.
Stack Overflow's 2025 developer survey found 84% of developers now use or plan to use AI tools, but positive sentiment toward them dropped from 70%+ in 2023-24 to 60% in 2025 (Stack Overflow). Adoption is up, enthusiasm is down. Familiarity breeds a sober read.
So the aggregate market for traditional software engineers is contracting. Individual productivity gains are narrower than the marketing. And within the contracting market, a specific kind of engineer is commanding premium comp.
The middle managers
Named examples, 2024–2026.
Amazon. September 2024: Andy Jassy set a target of raising IC-to-manager ratios by at least 15% by end of Q1 2025 (aboutamazon.com). Hit by end of March 2025, mostly through team consolidations and manager demotions rather than mass layoffs.
Google. Pichai told an all-hands on August 27, 2025 that Google had cut 35% of managers leading teams of fewer than three (CNBC). The ads division had its entire middle-management layer eliminated. Quote: "We need to be more efficient as we grow, so we don't just throw more people at every problem."
Microsoft. May 13, 2025: laid off about 6,000 people (3% of the company), with engineers as the largest category cut (817 at HQ, more than double product management). The stated goal in the Security division was to widen span of control from 5.5:1 to 10:1 engineer-to-manager (CNBC).
Meta. January 2025: 5% cuts targeting "low performers" with many laid-off staff reporting "At or Above Expectations" reviews (Fortune). About 25,000 cut since 2022. Operating margin expanded from 25% to 42% post-2023 cuts. Internal push toward "pod" structures for AI-assisted development.
Shopify. April 2025: Tobi Lütke's memo. "Reflexive AI usage is now a baseline expectation. Before asking for more headcount and resources, teams must demonstrate why they cannot get what they want done using AI" (Lütke on X; Fortune).
Salesforce. Benioff, February 2025: "Maybe we aren't going to hire any engineers this year." Salesforce cut 1,000+ roles in 2025 while adding 1,000–2,000 salespeople to sell AI agents (SF Standard).
Beyond the layoffs, the structural story is spans of control. Fewer managers. Larger teams. Korn Ferry's 2025 Workforce report found 41% of employees said their employer trimmed management layers in 2025. The traditional shape ("manager of 7–10, part of a 5-level chain") is being compressed, not pruned at the margins. That's a structural shift, right?
The AI engineer
In the middle of this, a new role shape took off.
AI Engineer. Builds applications on top of foundation models: prompt design, evals, RAG, fine-tuning, deploying LLM-backed features to production. Product-first. (Training models is still the ML Engineer role.)
The term was popularised by Chip Huyen's 2024 essay and her 2025 O'Reilly book of the same name. LinkedIn named "AI Engineer" the #1 "Jobs on the Rise" title for young workers two years running (LinkedIn Economic Graph).
Comp data, Levels.fyi Q3 2025:
- Anthropic software engineer: $563K (Senior) to $774K (Lead) total comp.
- Forward-deployed engineer at OpenAI / Anthropic: $350K–$550K, up to $700K total.
- At Staff level, the AI-engineer-vs-generic-SWE gap can exceed $400K (Intuit and others).
- Counterintuitive caveat: at entry level, the AI premium shrank from 10.7% in 2024 to 6.2% in 2025. At the bottom, "AI engineer" on a CV is no longer the ticket it was. At the top, the gap widened.
Sibling roles worth knowing:
- Forward-Deployed Engineer (FDE). Coined at Palantir in the early 2010s ("Delta"). Now standard at OpenAI (Colin Jarvis leads), Anthropic (Applied AI), Cohere, Datadog. Half-engineer, half-solution-architect, half-PM. FDE postings at AI labs grew roughly 800–1000% in 2025 (Pragmatic Engineer; Sundeep Teki).
- Product Engineer. Generalist who ships full features end-to-end: UX sense, full-stack, data instinct. Common at Linear, Vercel, Stripe, Ramp.
- Design Engineer. A product engineer whose gravity well is the UI. Ships in code, not Figma hand-off. $200K+ at Vercel, Stripe, Linear.
- Founding Engineer. YC-universe generalist: owns large product slices, talks to users, does infra and growth. Typical 2025 YC package: $150K–$250K cash + 1.5–2.0% equity.
The common thread: the premium is paid for shipping AI-powered product end-to-end, against real customers, in production. "AI on a CV" and LLM-wrapper side projects don't clear the bar anymore.
Lean teams and the "solo unicorn" meme
The extreme version of the story gets told with three names.
Cursor (Anysphere). Hit $100M ARR in January 2025 (the fastest any SaaS company has ever done it), crossed $500M by June, and $1B+ annualised by a November 2025 Series D at a $29.3B valuation. Team size went from ~40–60 in early 2025 to ~150 by August (TechCrunch). Revenue per employee that dwarfs traditional SaaS.
Midjourney. Reached $200M ARR with around 11 FTEs at peak leanness. Scaled to $500M ARR with 107 employees in 2025 ($4.7M revenue per head). No VC. No traditional marketing.
Pieter Levels. Genuinely solo operator. About $138,000 a month as of November 2025 (~$3M/year). Photo AI accounts for roughly 70% of revenue. Zero employees.
The "solo billion-dollar founder" is still more meme than norm. The underlying shift is real: revenue per employee is up sharply in AI-native product companies, and it's changing hiring upstream. Teams hire slowly and senior. Founding engineers with a demonstrated shipping record get recruited out of small teams at breakneck pace.
What's dying, what's growing
Contracting inside software:
- Entry-level backend at big cos.
- Manual QA (some of it absorbed by AI-assisted test generation).
- Junior frontend roles where the work is wiring up known patterns.
- Traditional "manager of 8 ICs, 5 levels of chain above" roles.
Growing:
- AI engineer (applied, product-facing).
- Forward-deployed engineer at AI labs.
- Product engineer and design engineer at Linear/Vercel-style companies.
- Evals / model-ops / AI infrastructure.
- Staff+ IC roles with some people leadership (tech-lead-manager), as managers move back toward shipping.
- Skilled trades (electricians +9%) for the data-centre build-out.
What to do about it
If you're mid-career in traditional software, the safe-seeming move is to stay where you are. I think that's no longer safe. Skilling into AI-applied work (RAG, evals, production LLM deployment) puts you on the right side of the split. Shipping real AI features at your current job is the cheapest way to prove that on a CV.
If you're in middle management, read the span-of-control data carefully. Spans are widening. Chains are compressing. A lot of former managers are moving back to staff-plus IC roles where they still do some people leadership. That's often a better trajectory than protecting a title that's being structurally deprecated.
If you're early-career, don't optimise for "AI engineer at a frontier lab" as a first job. Optimise for a team small enough that you ship end-to-end, see real users, and can point to a feature you owned. Founding-engineer and product-engineer roles at Series A startups give you more of that signal than a junior role at Google.
And if you're hunting: the roles that matter have hybrid titles that don't map to legacy "Senior Software Engineer" filters. Scoring CVs against role content rather than title strings is the whole point of Flint. The title search misses the job you actually want.
Try Flint free. 60 seconds, a CV, your top 10 scored matches across 100+ company boards.