TL;DR
Dice's July 2026 Tech Job Report says AI skills now appear in 75% of US tech postings — up from 73% in May, and +178% year over year. Overall postings are +3% month over month and +27% YoY, which Dice frames as "recovery into steady growth." Manufacturing led with +35% MoM. The fastest-growing skills YoY are Agentic AI (+587%) and AI Agents (+503%).
Useful. Also: not about you.
An aggregate report tells you the market moved. It does not tell you what your market — your city, your stack, your seniority band, your 40 target companies — is asking for this month. Nobody publishes that report, because your market is a slice of one. So you build it yourself, and it turns out job boards are one of the best scraping targets in existence: structured, paginated, public, and refreshed constantly.
ScrapeMaster does this in a side panel — run your search, let auto-detect find the listing rows in 2–4 seconds, use Follow detail to pull each full job description, export to CSV or XLSX, and count keywords in a spreadsheet. One afternoon to set up, ten minutes a month to maintain.
What Dice actually found
The July 2026 report, in numbers:
| Metric | Value |
|---|---|
| Tech postings mentioning AI skills | 75% (up from 73% in May) |
| AI skills mentions, YoY | +178% |
| Overall postings, MoM | +3% |
| Overall postings, YoY | +27% |
| Leading sector, MoM | Manufacturing, +35% |
| Fastest-growing skill, YoY | Agentic AI, +587% |
| Second fastest, YoY | AI Agents, +503% |
Dice's own framing for the trajectory is "recovery into steady growth" — not a boom, a recovery.
Three things stand out.
The 75% is a mention rate, not a requirement rate. A posting that says "familiarity with AI tools a plus" counts the same as one demanding three years of LLM fine-tuning. That's not a criticism of Dice — it's how keyword analysis works — but it's the single biggest reason the number can mislead you. A 75% mention rate is compatible with a world where 20% of jobs actually need AI skills and 55% of hiring managers added a line to look current.
Agentic AI at +587% is a base-rate artifact and still meaningful. Growing 587% from a small base is easy math. But the composition matters: the fastest-growing skills aren't "AI" generically, they're agentic specifically. That's a signal about which direction the demand is pointed, even if the percentage is inflated by where it started.
Manufacturing at +35% MoM is the sleeper. Everyone reads tech job reports for the tech-company numbers. The sector leading month-over-month growth isn't a tech company sector at all.
The counter-signal you should hold at the same time
Here's where most coverage of "AI skills in 75% of postings" goes wrong: it slides straight into "AI is taking the jobs."
On July 6, 2026, Microsoft cut roughly 4,800 roles — about 2.1% of its workforce. Xbox was hit hardest with 1,600. Xbox CEO Asha Sharma called it "the most significant restructure in Xbox history." Management layers went from 14 to 3–5, and four studios were divested. Roughly 3,200 more cuts are expected through FY2027.
And Microsoft explicitly stated the roles "are not being replaced by AI."
That's a restructure. Flattening management from 14 layers to 3–5 and divesting studios is an org-design decision, not an automation story. You can believe the company or not, but you can't cite the layoffs as evidence of AI substitution while ignoring the company saying it isn't.
So hold both:
- AI skills are appearing in more postings than ever (75%, +178% YoY).
- The biggest tech layoff of the month was explicitly not an AI substitution.
- Overall postings are up 27% YoY.
Both things are true at once. The market is asking for AI skills and growing and restructuring. That's a more interesting and more accurate picture than either headline alone. "AI skills are in demand" and "AI is eliminating roles" are different claims with different evidence, and conflating them produces bad career decisions — like panic-pivoting out of a specialty that's actually fine.
We've covered the layoff-tracking angle in more depth in scraping Meta and Microsoft layoff talent intelligence.
Why you need your own numbers
Dice's 75% is a national aggregate across all US tech postings. Ask yourself what you'd actually do with it.
You're a mid-level backend engineer in Denver targeting 40 fintech companies. Does 75% tell you:
- Whether Denver fintech backend postings mention AI? No.
- Whether they mean "we use Copilot" or "you'll build agents"? No.
- Whether your specific 40 targets are hiring at all? No.
- Whether the mention rate in your niche is rising or falling? No.
- What's actually in the job descriptions, as opposed to the titles? No.
The aggregate is a weather report for a continent. You need the forecast for your street. The gap between them isn't a small one — national tech hiring can be +27% YoY while your specific niche in your specific city is flat or down, and you'd never know from the headline.
The good news: this is one of the most tractable scraping jobs there is. Job boards are structured (every listing has the same fields), paginated (predictable), public (no login for search results on most boards), and fresh (they update constantly). It's close to a best case.
The workflow
Step 1 — Define your query and freeze it
Pick your search: role, location, seniority, remote/onsite. Write down the exact query string and the exact filters. You will run this identical search every month, and any change to it invalidates your comparison. Keep it in a note next to your spreadsheet.
Step 2 — Run the search and open the side panel
Run it on the board. Open ScrapeMaster from the toolbar; it opens in a Chrome side panel, so the listings stay visible while you work.
Step 3 — Let auto-detect find the rows
Auto-detect identifies the repeating listing pattern in 2–4 seconds and names the columns for you. No CSS selectors, no code. It reads the post-render DOM, so React-based boards work fine.
Step 4 — Keep the right columns
Trim to what you'll actually analyse:
- title
- company
- location
- date posted
- salary (if shown — many boards don't)
- listing URL
- date collected ← add this, always
Rename them to something legible. date_collected beats whatever the DOM called it.
Step 5 — Turn on pagination
Next-page buttons, load-more, numbered pagination, and infinite scroll are all handled. Set an extraction delay. This is not optional politeness — see the honesty section below. A few hundred listings at a civilised pace is fine; hammering a board is how you get blocked and how you deserve to.
Step 6 — Follow detail for the actual descriptions
This is the step that makes the whole thing work. The listing row gives you a title. The description gives you the skills — and the skills are the entire point.
Follow detail opens each listing's link in a background tab, extracts the fields you specify (full description text, requirements section, posted date), and merges them back into the row. Turn it on, point it at the description body, and let it run. This is slower than row extraction by design; give it delays and let it work.
Without this step you're counting words in job titles, which tells you almost nothing. With it, you have the actual text hiring managers wrote.
Step 7 — Export
CSV or XLSX for spreadsheet work, JSON if you're scripting, clipboard for a direct paste into Google Sheets. Everything's stored locally in IndexedDB until you export it.
Step 8 — Count the keywords
In a sheet, one column per skill term:
=IF(ISNUMBER(SEARCH("agentic", $F2)), 1, 0)
Terms worth counting, given what Dice found:
| Category | Terms to count |
|---|---|
| Generic AI | AI, artificial intelligence, machine learning, ML |
| Agentic (the growth area) | agentic, AI agent, agent framework, orchestration |
| LLM-specific | LLM, prompt, RAG, fine-tuning, embeddings |
| Tooling | Copilot, LangChain, vector database, MCP |
| Your stack | whatever you actually do |
Sum the columns, divide by total listings, and you have your AI-mention rate. Now compare it to Dice's 75%.
That comparison is the payoff. If your niche is at 40%, the national number is describing a market you're not in — and either that's an opportunity or a warning, but at least you know which conversation you're having. If your niche is at 90%, you're behind and the aggregate was understating your urgency.
Also count agentic separately from AI. Dice's +587% says that's where the movement is; your own data tells you whether it's arrived in your niche yet or is still a coastal rumour.
Step 9 — Build the time series
Same query, same board, same filters, once a month, keep the date column, append to one sheet. Never overwrite.
After three months you can see direction. After six you can see trend. ScrapeMaster saves per-domain extraction configs and re-applies them automatically, so month two takes about ten minutes — you run the search, open the panel, and the setup is already there.
The chart you're building — "% of postings in my niche mentioning agentic AI, by month" — does not exist anywhere else and cannot be bought. That's the point.
For recruiters and workforce planners
Same machinery, different question. Instead of "what should I learn," you're asking "what is everyone else requiring, and what are we requiring, and are those the same?"
- Competitive requirement analysis. Pull every posting for your target role across 15 competitors. Count skill mentions. Find where your JD is out of step.
- Salary transparency mapping. Where salary is disclosed, extract it. Coverage varies by jurisdiction; some states mandate it and you'll get real data, elsewhere it's sparse. Report the coverage rate honestly.
- Sector shifts. Dice says manufacturing led +35% MoM. Is that visible in your region? Only your own pull will say.
- Time-to-fill signals. Track which postings persist month over month. A role that's been open five months is telling you something about that company's comp band.
The same extraction also underpins shortlist-building, though that's a different post — and a different set of privacy obligations, since candidate data is personal data in a way that job-description text isn't.
The honesty section
We'd rather lose the install than have you get burned.
What ScrapeMaster can't do
- Cannot bypass login walls. If a board requires sign-in to see listings, you have to sign in yourself — and then it reads what your session can see. It does not get you past the wall.
- Cannot solve CAPTCHAs. There's no module. If you get challenged, you solve it as yourself.
- Does not rotate proxies or fingerprints. Your requests come from your IP because they are your requests. We're not an evasion tool and we're not going to become one.
- Heavy extraction on aggressive anti-bot sites can trigger blocks. LinkedIn and Cloudflare-protected boards in particular. This is a real, common outcome — not an edge case. Use extraction delays. If a site blocks you, it blocked you, and we can't fix it.
LinkedIn specifically
LinkedIn's terms prohibit scraping. We're not going to walk you around that, and we'd rather say so plainly than let you find out from a suspended account. We've written up the social media scraping rules and whether web scraping is legal if you want the full picture, including the CFAA and contract distinction that makes "public data" a more complicated defence than people assume.
Plenty of job boards have friendlier terms. Read them. It takes five minutes and it's the difference between a data project and an incident.
Personal data in job listings is regulated
This one gets skipped constantly. Job listings often contain recruiter names, email addresses, and phone numbers. That is personal data, and it is in scope for GDPR and CCPA regardless of the fact that someone posted it publicly. Public does not mean unregulated — that's the single most common misconception in scraping, and it's wrong in both regimes.
Concretely:
- If you don't need recruiter names for your skills analysis — and you don't — don't collect them. Drop the column. Purpose limitation isn't a formality.
- If you do collect them, you have a lawful basis to establish, retention limits to honour, and possibly notice obligations to the people concerned.
- "I'm just an individual job seeker" is not a blanket exemption, and the exemptions that might apply are narrower than people hope.
The skills analysis in this post needs title, company, location, date, and description text. None of that is personal data. Keep it that way — it's better analysis and less exposure. Our web scraping privacy compliance guide goes deeper. Not legal advice.
How this compares to the alternatives
| Approach | Cost | Your niche? | Your effort |
|---|---|---|---|
| Read the Dice report | Free | No — national aggregate | Ten minutes |
| Buy a labour-market data product | Thousands/year | Sometimes, if your slice is big enough | Low |
| Octoparse / ParseHub / Import.io | Paid tiers, cloud | Yes | Workflow building, real learning curve |
| Simplescraper / Web Scraper.io | Free tier + paid cloud | Yes | Recipe or sitemap setup |
| Instant Data Scraper / Thunderbit | Free / freemium | Yes | Low |
| ScrapeMaster | Free | Yes | One afternoon, then ten min/month |
The cloud tools are genuinely more powerful for continuous large-scale crawling — that's a real thing they do that we don't. But for "40 target companies, one query, once a month, from my own browser," a cloud crawl is the wrong shape entirely: more infrastructure, more cost, more terms-of-service exposure, and it doesn't see what your session sees.
If you want the step-by-step version aimed specifically at this task, see scraping job listings from job boards.
Keep the listings you actually applied to
A footnote that saves people real pain: postings get taken down. When you're six weeks into a process and want to check what the role originally said about scope, comp, or remote policy, the URL is often a 404.
Extract rows for analysis, but for the handful of jobs you actually apply to, save a dated PDF. Convert: Web to PDF captures it locally with selectable text and working links, no watermark, no upload. We wrote about saving job listings as PDF specifically for this. Different jobs: rows for counting, PDFs for keeping.
Frequently asked questions
Does 75% of postings mentioning AI mean 75% of jobs require AI skills?
No — and this is the most important caveat in the report. It's a mention rate. "Familiarity with AI tools a plus" counts identically to "3 years building agents." Pull the descriptions yourself and you can grade mentions by seriousness, which the aggregate cannot.
Is AI taking these jobs?
The Dice data doesn't say that, and the month's biggest counter-example points the other way: Microsoft cut ~4,800 roles on July 6, 2026 and explicitly said they "are not being replaced by AI" — it was a restructure, with management layers going from 14 to 3–5 and four studios divested. Meanwhile postings are +27% YoY. AI skills being in demand and AI eliminating roles are separate claims needing separate evidence.
Can ScrapeMaster scrape LinkedIn jobs?
LinkedIn's terms prohibit scraping, and their anti-bot systems are aggressive. We're not going to help you around either. Use boards with friendlier terms — there are many, and they're better structured anyway.
Will I get blocked?
Possibly, on aggressive anti-bot sites, especially with heavy extraction. We don't rotate proxies or fingerprints and can't solve CAPTCHAs, so if you're blocked, you're blocked. Use extraction delays, keep volume modest, and don't run Follow detail across 2,000 listings at full speed.
Is it legal to scrape job postings?
Scraping publicly-accessible data is generally legal in most jurisdictions, but that's the beginning of the analysis, not the end. Terms of service, copyright in the description text, and GDPR/CCPA obligations around recruiter personal data all apply independently. Not legal advice — see our legality guide.
Do I need to code?
No. Auto-detect finds the pattern in 2–4 seconds and names columns. The only "code" is a spreadsheet SEARCH() formula, and you can copy the one above.
How many listings do I need for a meaningful rate?
For a stable percentage, aim for 100+ per monthly pull. Below ~50 the number bounces on noise. If your niche is genuinely tiny, report the raw counts instead of a percentage — "7 of 12 postings mentioned agentic AI" is more honest than "58%."
Where does my data go?
Nowhere. Rows live in your browser's IndexedDB and export to CSV, XLSX, JSON, or the clipboard. The only network request is during auto-detect, when page HTML structure — not content — goes to our analysis API to suggest columns.
Bottom line
Dice's July 2026 report is real and worth reading: AI skills in 75% of US tech postings, +178% YoY, postings +27% YoY, Agentic AI +587%, manufacturing leading at +35% MoM. And on July 6, Microsoft cut ~4,800 roles while explicitly saying AI wasn't replacing them. Both are true. Anyone telling you only one of them is selling something.
What neither tells you is what your city, your stack, your seniority band, and your 40 target companies are asking for this month. That report doesn't exist. Build it: freeze a query, extract the rows, follow the detail links for the real descriptions, keep the date column, count the keywords, run it again next month. An afternoon to set up, ten minutes a month after that, and by October you'll have a chart nobody can sell you.
ScrapeMaster is free — no account, no trial, no upsell. Side panel, auto-detect in 2–4 seconds, pagination, Follow detail, CSV/XLSX/JSON/clipboard, everything local. It won't get you past a login wall or a CAPTCHA, and we're not going to pretend otherwise.