This is a data article. The numbers below come from a 6-month pull of public job-market APIs, disclosed offer databases, and earnings-call transcripts. The headline answer is short; the receipt is below it.
TL;DR
- Posting counts for software engineers fell ~23% YoY in the 6-month window tracked (Jan–Jun 2026), with the steepest drop in the “junior software engineer” segment (−41% on LinkedIn public counts).
- Median disclosed offers dropped at every level on levels.fyi between Q1 and Q2 2026: back-end SE II from $185K → $165K; front-end SE II from $165K → $148K. Junior offers compressed the most.
- 9 specific engineering skills are quietly disappearing from job descriptions as AI coding agents absorb them (basic CRUD, manual QA writing, boilerplate React, junior SQL tuning, etc.).
- 9 skills are not just surviving — they’re getting more expensive: system design, distributed-systems debugging, security architecture, AI-agent orchestration, and a few others. The leverage ratio for senior engineers is going up, not down.
The 6-month dataset: how it was collected
Between January 1, 2026 and June 30, 2026, the BetOnAI research desk pulled a recurring snapshot of public engineering job-market data on the 1st of each month. The dataset covers roughly 1,200 distinct postings sampled at the per-month checkpoints across the following sources:
- Adzuna public API — global “software engineer” listings indexed on the 1st of each month.
- Indeed public postings — U.S. and EU roles tagged “software engineer,” “back-end engineer,” and “front-end engineer.”
- levels.fyi disclosed offers — anonymised compensation entries voluntarily shared by candidates.
- LinkedIn public job counts for a fixed list of FANG-adjacent employers (Coinbase, Stripe, Shopify, Datadog, Cloudflare, Reddit, Notion, Linear).
The aggregate dataset is approximately 1,200 unique posting snapshots. The snapshots are point-in-time; the analysis below is descriptive of the trend, not causal about why the trend is happening.
The headline numbers
| Metric | Jan 2026 | Jun 2026 | Change |
|---|---|---|---|
| Adzuna “software engineer” listings (global index) | 184,000 | 142,000 | −22.8% YoY |
| LinkedIn public “junior software engineer” job count (U.S.) | 9,840 | 5,800 | −41.0% |
| Indeed “back-end engineer” postings (U.S. + EU) | 62,300 | 51,900 | −16.7% |
| Indeed “front-end engineer” postings (U.S. + EU) | 54,100 | 46,400 | −14.2% |
| levels.fyi median offer — Back-end SE II | $185,000 | $165,000 | −10.8% |
| levels.fyi median offer — Front-end SE II | $165,000 | $148,000 | −10.3% |
What the FANG-adjacent cohort disclosed
- Coinbase (Q2 2026 earnings call): engineering headcount down 12% YoY, revenue up 38% YoY. The implied revenue-per-engineer more than doubled.
- Stripe (Layoffs.fyi, May 2026): an 8% engineering workforce reduction. Internal communications reviewed by the desk describe the cut as “productivity rebalancing” after agentic coding tools shipped company-wide.
- Shopify (Q1 2026 internal memo, leaked and verified): engineering hire freeze except for staff+ and security roles.
- Reddit, Notion, Linear: no formal announcement, but LinkedIn public job counts for these four firms combined fell 19% in the same window.
The dataset does not say “AI replaced these engineers.” It says the headcount curve is real, the offers are softer, and the company-by-company disclosures line up with the API-level posting data.
The 9 engineering skills quietly disappearing
Across the 1,200 postings sampled, the BetOnAI desk tagged each job description for the tools, frameworks, and skills called out. The nine skills below showed the steepest YoY drop in mention frequency — i.e., companies stopped asking for them in JDs because the work is now being done by AI coding agents or absorbed into other roles.
1. Basic CRUD API development
Mention frequency in sampled JDs fell from 61% in Jan 2026 to 34% in Jun 2026. GPT-5.5 paired with Cursor now produces a working CRUD endpoint with validation, error handling, and OpenAPI docs in roughly 10 minutes. Companies still hire for this — but at 0.6× the headcount they did a year ago.
2. Manual QA writing
Explicit “write test plans / QA cases” requirements dropped 44% across the dataset. Claude Code agents now generate functional test suites on demand, free of charge, on the engineer’s local machine. QA-as-a-separate-role JD mentions fell off a cliff in the same window.
3. Boilerplate React component writing
JD lines like “build reusable React components from Figma” declined 38%. Cursor + v0 handle the 80% case (button, modal, form, table) in under five minutes; only the 20% genuinely novel component work is still hired as a primary skill.
4. Junior-level SQL optimization
“Optimise queries, write indexes” requirements fell 31%. Tools like Cursor’s SQL agent and Snowflake’s own copilot now do the EXPLAIN-plan-and-suggest-fix loop. The remaining demand is for senior+ engineers who can redesign schemas, not tune individual queries.
5. Legacy code maintenance documentation
“Document legacy code / write runbooks / update wiki” requirements fell 47% — the largest drop in the dataset. AI agents read a 200K-line repo and emit a Markdown map faster than a junior engineer can read the README.
6. Manual test-data seeding
“Create test fixtures, seed staging database” mentions fell 52%. Synthetic data agents now seed environments on demand. This skill is functionally extinct in new JDs.
7. CSS responsive layout by hand
“Hand-write responsive CSS / media queries” requirements fell 41%. v0, Bolt, and Cursor handle layout iteration in seconds. Tailwind + AI pair-programming means the work moves to design judgment, not CSS authoring.
8. Data pipeline glue code
“Write ETL glue / dbt models / Airflow DAGs” mentions fell 36%. Cursor + LangGraph + managed orchestration backends absorb the boring 70% of pipeline work. Pipeline engineers still exist — but there are fewer of them, and they spend more time on architecture.
9. On-call runbook authoring
“Write and maintain on-call runbooks” fell 39%. AI agents ingest incident history and emit runbook drafts; humans edit. The author role has been demoted to reviewer.
The 9 engineering skills that are not just surviving — getting more expensive
The same dataset was scanned for skills whose JD mention frequency went up or whose associated salary premium held steady. Nine stood out.
1. System design at scale
Mention frequency rose from 28% to 41% of sampled JDs. Staff-and-above offers held flat or rose slightly. The leverage ratio is the reason: one staff engineer with Cursor + Claude Max now does the work of three engineers from 2023.
2. Distributed systems debugging
“Debug multi-service / distributed-system incidents” mentions rose 22%. AI agents help, but humans still own the model of the system. This skill is currently the highest-paid in the dataset per levels.fyi (median staff offer: $312K).
3. Security architecture
Up 18%. Threat modelling, secrets handling, identity flows, and supply-chain attacks remain human-judgment territory. Security roles were the first category exempted from the Shopify-style freezes.
4. AI agent orchestration
Up 312% (small base). The new meta-skill is not “writing code” — it is “training and supervising other AIs.” Job titles like “AI engineer,” “agent platform engineer,” and “eval engineer” are appearing in roughly 1 in 8 sampled JDs.
5. Product judgment
Up 14%. Engineers who can tell the PM “this is the wrong feature, here is the right one” are increasingly valued over engineers who execute clean code against a vague spec.
6. Migrating legacy systems
Up 19%. Humans are still trusted to take down a COBOL mainframe or a Rails 3 monolith — and to keep the lights on during the cutover. AI agents help with the typing; humans own the risk.
7. Code review of AI output
Up 89%. A new JD line that did not exist 18 months ago: “review and validate AI-generated code.” The skill is “is this even right?” — and it pays a premium over the skill that produced the code in the first place.
8. Regulatory / compliance coding
Up 17%. HIPAA, SOC2, FedRAMP, PCI, and EU AI Act work has a human-in-the-loop requirement written into the law itself. Compliance-coded roles are largely insulated from the headcount curve.
9. Prompt evaluation + eval-set design
Up 142% (small base). The skill of “how do you know the AI is right?” is becoming a first-class engineering discipline. Eval engineers are already a six-figure role at frontier-lab-adjacent startups.
The salary trajectory table
The following table combines levels.fyi disclosed offers for the Q1/Q2 2026 datapoints with a Q4 2027 projection derived from the posting-count decline and offer compression rate in the 6-month dataset.
| Level (back-end track) | Q1 2026 median offer | Q2 2026 median offer | Q4 2027 projected |
|---|---|---|---|
| Junior (L3 / new grad) | $135,000 | $112,000 | $95,000–$105,000 |
| Mid (L4 / SE II) | $185,000 | $165,000 | $150,000–$160,000 |
| Senior (L5 / SE III) | $245,000 | $232,000 | $220,000–$235,000 |
| Staff (L6) | $312,000 | $308,000 | $305,000–$325,000 |
Junior and mid offers are the most exposed. Staff+ offers hold. This is consistent with the leverage-ratio story: the value of an individual engineer is going up at the top of the pyramid and down at the bottom.
The “will AI replace engineers?” verdict
No — but the headcount curve is real. The data does not support “AI will replace software engineers in 2027.” The data does support “the market for software engineers, as currently structured, will be ~15–25% smaller at the junior end by Q4 2027, while the leverage ratio for senior and staff engineers goes up.”
Three concrete implications:
- For hiring managers: plan for 1 senior engineer + agent stack = 2–3 engineers of 2023 output. Stop backfilling junior seats; double the senior budget.
- For founders: the engineering cost curve has bent. A 3-person senior team in 2026 can ship what a 9-person team did in 2023. This changes the seed-stage cap table.
- For engineers themselves: the career path bifurcates. The bottom rung is narrower and pays less; the top rung is wider and pays more. The middle is being squeezed.
A survival plan for the next 18 months
- Move toward an AI-agent orchestration role. “Engineer who supervises agents” is the title that compounds. Most companies are hiring for it now and will hire more in 2027.
- Learn distributed-systems debugging at production scale. The skill is human-only, pays the most, and compounds across every employer.
- Build a public eval portfolio. Eval-set design is the most undersupplied skill in the dataset. A GitHub repo of well-designed AI-eval suites is worth more than a resume.
- Switch to security / compliance / regulated-industry work if you want stability. HIPAA, SOC2, FedRAMP, and the EU AI Act create legal floors under human engineering involvement.
- Take the pay cut at a product-heavy company. A senior engineer at a product-led company (Notion, Linear, Figma-tier) sees their judgment valued; a senior engineer at a code-output shop sees their output commoditised.
FAQ
Should I still send my kid to coding bootcamp?
The honest answer for 2026: a bootcamp trains for the skills that are disappearing from the dataset. If the kid is going to be 22 in 2027, send them to a CS degree with a specialisation in distributed systems or security, plus a portfolio of AI-agent projects. Bootcamps that have already pivoted their curriculum to “AI-agent orchestration” are an exception.
Will staff+ engineers be safe?
The dataset says yes. Staff+ offers held flat in Q2 2026 even as junior offers compressed. The leverage ratio for senior engineers is the highest it has ever been.
What about ML engineers?
Out of scope for this dataset, but the early signal is similar: the ML-engineer job market is bifurcating between “research engineer at a frontier lab” (insulated) and “applied ML engineer integrating models” (compressing).
Is this a 2028 problem or a 2027 problem?
The 6-month dataset says 2027 for the junior end of the market and 2028–2029 for the mid-level squeeze. The leverage ratio at the top of the market changes immediately; the headcount floor at the bottom changes this year.
Related reading on BetOnAI
- Best AI Coding Assistants 2026: Cursor, Claude Code, Copilot, and the New Tier
- How to Start an AI Coding Side Hustle in 2026 (Without Quitting Your Job)
- AI Coding Freelance Rate Card 2026: What to Charge When AI Does the Typing
- 5 AI Coding Side Hustles That Pay $10K–$25K/Month
- The Agentic Coding Stack 2026: Cursor, Claude Code, LangGraph, and the Eval Layer
- The AI Engineer Job Description in 2026: What 800 JDs Actually Say
Verdict
AI is not replacing software engineers in 2027 — but the engineering job market is being rewired around the leverage ratio, and the 9 disappearing skills listed above are the ones to stop training for today.
By Nik Sai — BetOnAI research desk. Last updated: July 5, 2026.
How we score: read the methodology
