Which Jobs Are Worth Pursuing in the AI Era? Data First, Then the Map
We scanned the job descriptions of every active role in our database — the numbers at the top are read live from the job database and refresh hourly. For international students who need work sponsorship, roles that explicitly state no sponsorship need no discussion: however strong your AI skills, they cannot buy visa eligibility. What is worth watching is which JDs in the explicitly-sponsoring market carry clear AI technology signals, and how mention rates differ between entry-level and mid-level roles. Full-text keywords can also come from product descriptions or company introductions, so what we present is a JD mention rate, not a candidate-skill requirement rate.
Read live from the job database · refreshed hourly · full-text keyword scanning may include job requirements, product descriptions, or company introductions — not a candidate-skill requirement rate
Why two axes: AI narrows the entrance, visas gate the exit
Looking only at "will AI replace me" leads to wrong conclusions. The Stanford Digital Economy Lab's research on ADP payroll data finds: employment for 22–25-year-olds in the most AI-exposed occupations runs about 19% below trend — yet the paper's first finding is precisely that it finds no evidence of global, AI-induced employment replacement. The loss is concentrated on the hiring side: the door has narrowed, the building has not collapsed.
Meanwhile, entry-level roles face a second squeeze for international students: the H-1B lottery is now weighted by wage level (DHS's own estimate for the final rule: roughly 15.3% selection for Level I, about 61.2% for Level IV), and entry salaries naturally sit in the worst bracket; the UK Skilled Worker general threshold of £41,700 likewise keeps many entry roles out. Global postings for 0–2 years of experience are down about 29% since the start of 2024 (Randstad 2026 Workmonitor).
So this map asks only two questions: is this job safe under AI impact? Does it sponsor a work visa? Stack the two questions, and you get the job map international students should be reading.
The quadrant map
Axis definitions: the horizontal axis is entry-level safety under AI impact (based on posting trends and AI-substitution evidence); the vertical axis is visa sponsorship friendliness (based on whether the salary band clears visa thresholds, plus actual employer sponsorship behavior). One core pattern: the safer a job is from AI, the higher its salary band tends to be, and the more visa-friendly — the two axes often move together.
- AI / applied AI engineers — #1 on LinkedIn's 2026 US growth list, with a median requirement of just 3.7 years of experience
- Forward Deployed Engineer — median base around $190K–$200K; a high value-per-effort route for international students strong at both engineering and communication
- Data & AI infrastructure: Data Engineer, MLOps
- Cloud / DevOps, cybersecurity
- Domain-expert × AI hybrid roles: financial risk, healthcare AI, legal AI
High salary band → naturally friendly to visa thresholds; the best cell combining growth and accessibility.
- SDR / BDR (sales development) — up 17% against the trend in the UK, but visa sponsorship is generally unfriendly
- Localized retail-management roles
The roles are growing, but confirm sponsorship intent before investing time.
- Junior generic software development without an AI direction — UK entry-level postings −27%, and mostly in the H-1B Level I lottery bracket
Not "never apply" — just stop treating it as the default option; this cell is exactly where AI substitution and visa disadvantage stack.
- Basic translation and localization (#1 on Microsoft's AI-applicability list)
- Data entry, Tier-1 customer support, telesales (postings −25% to −34%)
- Entry-level accounting and audit (UK entry −29%)
- Template-style junior graphic design (UK −28%)
Cell by cell: where the safety and sponsorship signals come from
The quadrant ratings combine three kinds of evidence: posting trends (UK government × LinkedIn statistics on 38 entry-level occupations, April 2026), empirical signals of AI substitution (employer-side surveys and ATS behavior), and visa-system thresholds (the H-1B wage-weighted lottery, the UK Skilled Worker threshold). The table below shows the key data for representative roles in each cell.
| Role | Demand signal | Experience bar | Sponsorship signal |
|---|---|---|---|
| AI / applied AI engineer | #1 on LinkedIn's 2026 growth list | Median 3.7 years | Core sponsorship bracket among tech employers |
| Forward Deployed Engineer | New role created by agent deployment | Junior from $137K | High-pay band → friendly to visa thresholds |
| Data / MLOps engineer | 42% of large enterprises already run AI agents in production (Mayfield CIO survey) | Mostly mid-level | High-pay band |
| Junior generic software development (no AI direction) | UK entry postings −27% | Entry | Sponsored, but mostly in the Level I lottery bracket (~15.3%) |
| Entry-level accounting & audit | UK entry −29% | Entry | Removed from the UK TSL list |
Measured by track: the AI demand gap is wide
Below are live statistics by track (strict-tier JD mention methodology, refreshed hourly with the job database; covering only jobs already assigned to tracks — one job can belong to multiple tracks). The last column is the sponsorship-focused view — the AI technology mention rate among explicitly sponsoring jobs: for international students who need a work visa, this is the column to watch.
| Track | Active roles (in database) | JDs mentioning AI (strict tier) | Mention rate among sponsoring roles |
|---|---|---|---|
| Quant | 1,573 | 38.1% | 33.5% (472 sponsored roles) |
| Software engineering | 21,482 | 31.5% | 38.8% (6,178 sponsored roles) |
| Product management | 1,128 | 28.6% | 27.9% (387 sponsored roles) |
| Consulting | 4,176 | 17.7% | 22% (1,572 sponsored roles) |
| Investment banking / finance | 6,929 | 11.8% | 12.2% (3,456 sponsored roles) |
Read live from the job database · refreshed hourly · data as of 2026-09-27 · covers only jobs already assigned to tracks (one job can belong to multiple tracks; jobs outside tracks are not in this site's five product tracks)
Which skills exactly: the AI skill board (JD mention rate)
The previous table answers "which tracks mention AI technology more often"; this one answers "what exactly they mention". Three methodology notes before reading: ① this is a JD mention rate — keyword hits from scanning the full text of job descriptions, deduplicated per job — not a requirement rate, and a small share of hits come from product or company-introduction context (the watch-list notes name which terms carry this risk); ② different skills may overlap — one job can hit multiple items, so items cannot be summed; ③ the whole-database and explicitly-sponsoring readings sit side by side, and items with a sample under 30 show counts only.
① AI capabilities & methods board
| Skill | Mention rate (whole database) | Mention rate (sponsoring market) |
|---|---|---|
| Generative AI / LLMs (incl. GPT, ChatGPT variants; deduplicated per job) | 6.6% (5,003) | 10.2% (2,185) |
| Machine learning | 6.1% (4,621) | 7.3% (1,571) |
| AI governance (responsible AI / governance / safety) | 2.2% (1,652) | 3.0% (637) |
| Computer vision | 1.8% (1,386) | 0.8% (164) |
| Deep learning | 1.1% (869) | 1.3% (282) |
| RAG (retrieval-augmented generation) | 1.1% (841) | 1.3% (286) |
| NLP | 1.0% (735) | 1.6% (340) |
| Prompt engineering | 1.0% (729) | 1.1% (246) |
| Fine-tuning (LoRA / PEFT) | 0.7% (515) | 0.9% (198) |
| Reinforcement learning (RL / RLHF) | 0.6% (467) | 1.0% (214) |
| AI evaluation (LLM / model evaluation) | 0.3% (244) | 0.3% (68) |
| MLOps / LLMOps | 0.5% (381) | 0.6% (119) |
| Model deployment (serving / deployment / inference optimization) | 0.4% (335) | 0.4% (89) |
② Frameworks & platforms board
| Skill | Mention rate (whole database) | Mention rate (sponsoring market) |
|---|---|---|
| PyTorch | 1.3% (977) | 0.9% (197) |
| TensorFlow | 0.8% (593) | 0.6% (119) |
| GitHub Copilot | 0.3% (251) | 0.5% (102) |
| Agent / RAG frameworks (LangChain / LangGraph / LlamaIndex) | 0.5% (363) | 0.6% (130) |
| Cloud AI platforms (Azure OpenAI / SageMaker / Bedrock / Vertex) | 0.4% (328) | 1.0% (215) |
| scikit-learn | 0.3% (240) | 0.3% (62) |
| JAX | 0.3% (256) | 0.3% (60) |
| Hugging Face | 0.1% (110) | 0.2% (33) |
Watch list (below the main-board bar; readings carry context notes)
- AI agents / agentic: whole database 3,088 · explicitly sponsoring 901 — The signal is large and real, but hits include company-product context (spot-check precision below the 80% line); readings skew high
- GPT / ChatGPT (a sub-item of the row above, tracked for market wording): whole database 642 · explicitly sponsoring 335
- Microsoft 365 Copilot: whole database 53 · explicitly sponsoring 43 — Largely Microsoft-ecosystem sales / solutions job context (a different thing from the coding tool GitHub Copilot)
- Vector databases (Pinecone / pgvector / FAISS, etc.): whole database 70 · explicitly sponsoring 24 — Sponsored-listing sample < 30
- MLflow / Kubeflow / W&B: whole database 104 · explicitly sponsoring 22 — Sponsored-listing sample < 30
- Keras: whole database 62 · explicitly sponsoring 22 — Sponsored-listing sample < 30
Snapshot · data as of 2026-09-26 (live query unavailable; self-heals on the next refresh) · Methodology = JD mention rate (not a requirement rate — full-text hits include some product/company context) · items cannot be summed · items with samples <30 show counts only · main-board bar = spot-check precision ≥80% and sponsored sample ≥30
The most dangerous cell, once more on its own
"Junior generic software development without an AI direction" deserves its own emphasis, because it is the default path your target readers choose most often: study CS, apply for junior dev roles. The problem with this cell is the double squeeze — UK government × LinkedIn data shows entry-level postings for this occupation down 27% year over year (caveats: that report is correlational evidence and overall UK hiring was down 14% over the same period — read the number with both caveats in mind); and in the US it mostly sits in the H-1B Level I lottery bracket, where the selection rate is roughly halved to about 15.3%.
The way out is not to abandon engineering but to switch cells: the same engineering ability, plus an AI direction (AI / applied AI engineer, data & AI infrastructure) or plus a domain (Forward Deployed, industry-AI hybrid roles), moves you into the "worth pursuing" cell.
What is happening to entry-level jobs (2026 update)
Two external pieces of evidence pin down the background first. The Stanford Digital Economy Lab's tracking study (August 2026 revision, ADP payroll data): the employment gap for 22–25-year-olds in highly AI-exposed occupations has widened to 19% (prior reading 15%) — and the gap comes mainly from less hiring, not increased layoffs; the authors also state clearly that this is descriptive evidence, not a causal conclusion.
PwC's 2026 Global AI Jobs Barometer (2026-06-15, based on 1B+ job ads) sees the other side of the same coin from the employer side: entry-level jobs with the highest AI exposure are 7 times more likely than other entry-level jobs to also demand "senior-type" human skills such as leadership, judgment, and face-to-face communication; since 2019, these "professionalised" entry-level jobs have grown 35%, while the rest shrank 10%. Entry-level jobs are not disappearing — they are being redefined.
Mapped onto our job database — live readings refreshed hourly are in the table below, comparing the JD technology mention rates of entry-level and mid-level roles:
| Methodology (JDs mentioning AI technology · strict tier) | Entry-level | Mid-level |
|---|---|---|
| Whole database | 27.2% (2,812 / 10,333 roles) | 17.7% (11,316 / 63,764 roles) |
| Explicitly sponsoring market | 27.8% (1,249 / 4,495 roles) | 22.7% (3,631 / 16,015 roles) |
Read live from the job database · refreshed hourly · data as of 2026-09-27 · "Explicitly sponsoring" = the JD states sponsorship; it reflects employer intent and is not a guarantee for any specific role
What this means: not dodging AI, but turning it into a relative advantage
Three sources piece it together: the total number of entry-level jobs is contracting (Stanford), the entry-level jobs that remain are being "professionalised" (PwC), and in the current reading of our JD technology-signal mention rates, the entry-level figure is no lower than mid-level (table above — readings move with the live composition of the database). For international students who need a work visa, the effective response is not to avoid AI jobs, but to schedule the three capability layers — use AI, verify AI, and keep evidence AI cannot take away — into your time at school; the full layout is on the roadmap page (second button at the top of this page).
Non-technical backgrounds have an edge: the fastest growth is in "domain experts who use AI"
PwC UK's job data: AI "user" roles grew +65.8%, while "builder" roles grew only +21.6%. Indeed's six-country data: 63% of AI-tagged jobs sit outside the tech industry. The growth engine is not "the people who build AI" but "the people who use AI well inside their own domain".
This is a real advantage for international students in finance, biology, law, and design: no need to switch to code. Your domain knowledge plus the ability to use AI is a deeper moat than a pure-CS junior candidate's. For how to build the AI capability, see the roadmap page (second button at the top of this page).
After the offer: STEM OPT eligibility for AI roles and the E-Verify check
The map above answers "what to apply for"; this section answers "how the visa side lands". Post-completion OPT for F-1 students runs 12 months; if your degree is on the current DHS STEM Designated Degree Program List you can apply for a 24-month extension — 36 months in total (USCIS). Whether a degree counts as STEM is not decided by the program name: what matters is whether the CIP code on your I-20 appears on the list. For AI-track job seekers this is usually good news: on the current list (ICE, updated July 22, 2024), Computer Science (11.0701), Artificial Intelligence (11.0102), and Data Science (30.7001) are all included, and so are "domain × data" codes such as Business Analytics (30.7102) and Bioinformatics (26.1103) — which lines up with the "domain experts who use AI" route in the previous section.
On the employer side there are two hard requirements: the STEM OPT employer must already be enrolled in E-Verify (evidenced by a valid E-Verify company identification number) and must sign Form I-983, the training plan, with you (USCIS). Enrollment can be checked in advance — USCIS’s E-Verify Employer Search is a public lookup showing whether an employer is currently enrolled (refreshed daily at around 2 a.m. ET). Two honesty caveats: an employer may be enrolled under its legal name, so searching by the brand name can miss it; and an employer with multiple hiring sites can enroll only some of them — a missing result does not mean no sponsorship; confirm with the employer once you have the offer.
Unemployment days are the other red line: up to 90 cumulative days during OPT, plus 60 more on the STEM extension — 150 in total (USCIS). Filing windows, EAD start and end dates, and how to budget unemployment days — enter your graduation month and this site’s visa countdown tool computes it in one pass (UK Graduate visa and US OPT/STEM OPT side by side, checked against official sources line by line):
Methodology, caveats and disclaimers
- Top stats band and track table: both read live from this site's job database (refreshed hourly; the track table covers only jobs already assigned to tracks — one job can belong to multiple tracks, and jobs outside tracks are not in this site's five product tracks).
- "JDs mentioning AI technology" = strict-tier full-text keyword methodology (see FAQ #1); it can miss implicit demand and can include product or company-introduction context — it is not a candidate-skill requirement rate.
- Visa methodology: visa_sponsorship is read from what the JD explicitly states (sponsors / does not sponsor / unstated — all three states shown); "explicitly sponsoring" is not a guarantee for any specific role.
- External research figures (Stanford / UK government × LinkedIn / PwC / Randstad / Mayfield / LinkedIn Jobs on the Rise) are quoted from verified entries in this site's internal research ledger; the UK government × LinkedIn report is correlational evidence and must carry the "overall UK hiring down 14% over the same period" caveat.
- This page is a compilation of public information and data snapshots, not legal advice, and not investment or immigration advice; jobs and policies change at any time — for material decisions defer to official publications.
FAQ
How is "JDs mentioning AI technology (strict tier)" computed?
We scan the full text of every active job description and count a strict-tier signal when it hits technology terms such as machine learning / artificial intelligence / LLM / generative AI / GenAI / GPT / deep learning / NLP / Copilot / prompt engineering. Full-text keywords cannot perfectly separate candidate requirements from product descriptions and company introductions, so this is a JD mention rate, not a candidate-skill requirement rate; it can both miss implicit demand and include business context. The stats band at the top refreshes hourly with the job database.
Why might your AI technology mention rate be higher than numbers I see elsewhere?
Different methodologies — every number is telling the truth. LinkedIn's official blog (published December 2024, covering the 2023-24 window) reports member-side data: users adding the AI literacy skill grew 177% in a year, about five times the growth of all skills combined (36%) over the same period — a self-reported signal from job seekers and practitioners. Ours is a different methodology: we scan the full text of each job description for keywords (strict tier; see the live reading in the stats band at the top), hitting wherever the term appears — job requirements, product descriptions, or company introductions. So our number should be read as a JD technology-signal mention rate and cannot be swapped directly for member growth rates or label-only statistics.
Why is the AI mention rate for entry-level jobs even higher than for mid-level ones?
Two layers. ① Current reading: the live numbers in the top stats band and in this page's entry-level section (refreshed hourly) show the same shape — entry-level JDs mention AI technology signals at least as often as mid-level ones. ② Composition effect: at the 2026-09-13 snapshot we did a one-time within-track decomposition — looking only inside software engineering, entry 35.2% vs mid-level 28.8%, consistent with the whole-database direction at that time, showing the shape is not an artifact of entry-level jobs being concentrated in technical tracks. Live readings fluctuate as the database composition changes; that decomposition is a one-time verification at a historical point in time; full-text hits still do not equal a candidate-skill requirement rate.
Why focus on "explicitly sponsoring" jobs? What about non-sponsoring and unmentioned ones?
Roles that explicitly state no sponsorship are a hard exclusion for international students who need a work visa — however strong your AI skills, you cannot get past the visa gate. So the skill conclusions on this page take explicitly-sponsoring jobs as their baseline (fourth cell of the stats band at the top). Two reminders: "explicitly sponsoring" reflects employer history and intent, not a guarantee for you; a JD that does not mention visas does not mean no sponsorship — confirm role by role.
How often is this page updated?
The top stats band and the track table are both read live from the job database (refreshed hourly; the track table covers only jobs already assigned to tracks — one job can belong to multiple tracks, and jobs outside tracks are not in this site's five product tracks). The job database itself syncs daily from multiple ATS sources. If the track table temporarily shows a snapshot (the page marks the as-of date), the live query is temporarily unavailable and will self-heal on the next refresh.
Data & Sources
- Our job database · all active roles (synced daily from multiple ATS sources; the data source for this page's live stats band)
- Stanford Digital Economy Laboratory · entry-level employment and AI research (ADP payroll data)
- UK Government × LinkedIn · Entry-level opportunities in the AI era (April 2026; entry-level trends across 38 occupations)
- PwC · AI Jobs Barometer (annual report; "users vs builders" job growth rates)
- Federal Register · H-1B wage-weighted lottery final rule (Level I–IV selection rates are DHS's own estimates)
- USCIS · official STEM OPT extension page (24-month extension, E-Verify employer requirement, Form I-983, unemployment-day caps)
- ICE · DHS STEM Designated Degree Program List (current version July 22, 2024; CIP codes listed one by one)
- E-Verify Employer Search · public lookup of employer E-Verify enrollment (USCIS, refreshed daily)
Figures on this page are job-database snapshots and keyword-methodology statistics; live numbers are those in the page stats band. External research figures are quoted from public reports and remain the conclusions of their original authors. This page is not legal, immigration, or career advice; visa and hiring policies may change at any time — defer to official publications for personal decisions, and consult a licensed professional when needed.