Skills AI engineer job postings ask for, ranked
Python is the default, and the new default is LLMs. Of 4,859 US postings for AI and ML engineering roles, 57.9% name Python in the text we read and 48.4% name LLMs. PyTorch appears in 22.1%, nearly twice as often as TensorFlow (11.4%).
By OfferMagnet · · 5 min read
The short answer
- name Python (2,815 of 4,859 postings); 91.8% of the 3,067 that name any programming language
- 57.9%
- name LLMs or large language models (2,354); 40.4% mention AI agents (1,961)
- 48.4%
- name PyTorch (1,072), against 11.4% for TensorFlow (555) and 6.0% for JAX (291)
- 22.1%
The top of the list is the stack you would guess. The surprise is how much of it is now LLM work: LLMs, agents, RAG (18.3%) and evals (18.6%) all rank above Kubernetes (13.1%) and SQL (9.7%). Classic cloud and infrastructure skills show up more as the title gets more senior. Senior, staff and principal postings name Kubernetes in 14.8% of cases against 11.5% for the rest, and Spark in 10.5% against 7.4%.
Our counts are a floor. Many postings never list tools, and we read only the first 4,000 characters of each.
The top 25, split by seniority
| Skill | Postings | All | Mid or unspecified (2,373) | Senior and up (2,433) |
|---|---|---|---|---|
| Python | 2,815 | 57.9% | 60.7% | 55.1% |
| LLMs | 2,354 | 48.4% | 45.3% | 51.6% |
| AI agents | 1,961 | 40.4% | 37.9% | 42.3% |
| Distributed systems | 1,164 | 24.0% | 21.4% | 26.6% |
| PyTorch | 1,072 | 22.1% | 24.4% | 19.9% |
| AWS | 1,057 | 21.8% | 20.6% | 23.0% |
| C++ | 1,014 | 20.9% | 22.7% | 19.1% |
| Evals | 905 | 18.6% | 19.6% | 17.8% |
| RAG | 889 | 18.3% | 16.6% | 20.1% |
| Java | 815 | 16.8% | 17.1% | 16.4% |
| MLOps and model serving | 709 | 14.6% | 12.5% | 16.8% |
| Kubernetes | 636 | 13.1% | 11.5% | 14.8% |
| GCP | 625 | 12.9% | 12.3% | 13.7% |
| Fine-tuning | 622 | 12.8% | 13.2% | 12.5% |
| Azure | 603 | 12.4% | 11.9% | 13.0% |
| NLP | 582 | 12.0% | 11.5% | 12.2% |
| TensorFlow | 555 | 11.4% | 11.8% | 11.1% |
| SQL | 470 | 9.7% | 9.2% | 9.7% |
| Vector databases | 445 | 9.2% | 8.6% | 9.8% |
| Spark | 437 | 9.0% | 7.4% | 10.5% |
| LangChain or LangGraph | 403 | 8.3% | 7.7% | 9.0% |
| Reinforcement learning | 382 | 7.9% | 9.3% | 6.5% |
| Prompt engineering | 376 | 7.7% | 7.2% | 8.3% |
| Docker | 372 | 7.7% | 7.5% | 7.9% |
| TypeScript | 371 | 7.6% | 7.0% | 8.3% |
Seniority barely changes the order. The pattern that does show up: mid-level postings lean on PyTorch (24.4% vs 19.9%) and reinforcement learning (9.3% vs 6.5%), which suggests more research-adjacent roles. Senior postings add platform words: MLOps, Kubernetes, Spark, distributed systems.
Languages and frameworks
| Name | Postings | Share |
|---|---|---|
| Python | 2,815 | 57.9% |
| C++ | 1,014 | 20.9% |
| Java | 815 | 16.8% |
| TypeScript | 371 | 7.6% |
| Go | 308 | 6.3% |
| JavaScript | 217 | 4.5% |
| Rust | 137 | 2.8% |
| PyTorch | 1,072 | 22.1% |
| TensorFlow | 555 | 11.4% |
| JAX | 291 | 6.0% |
| CUDA | 182 | 3.7% |
| Triton, vLLM or TensorRT | 234 | 4.8% |
Python is close to universal wherever a language is named: 91.8% of the 3,067 postings that name Python, C++, Java, Go, Rust or TypeScript name Python. C++ is the clear second, driven by inference, kernels and robotics roles. Rust and CUDA are rare in absolute terms (2.8% and 3.7%) but both skew to mid-level titles.
What travels together
| Pair | Postings with both | Share of all | Second skill, given the first |
|---|---|---|---|
| LLMs + Python | 1,493 | 30.7% | 63% of LLM postings name Python |
| AI agents + LLMs | 1,330 | 27.4% | 68% of agent postings name LLMs |
| AI agents + Python | 1,102 | 22.7% | 56% of agent postings name Python |
| C++ + Python | 889 | 18.3% | 88% of C++ postings name Python |
| PyTorch + Python | 843 | 17.3% | 79% of PyTorch postings name Python |
RAG is nearly always an LLM word: 88% of RAG postings (778 of 889) also name LLMs. Evals travel the same way, with 70% of eval postings naming LLMs.
What we counted
OfferMagnet reads postings from company careers pages and job boards. We kept US postings first seen between 19 September and 30 October 2026 (77,814) and found the AI and ML engineering ones by title.
- AI-titled: the title contains AI, ML, LLM, GenAI, generative AI, machine learning, applied scientist or research engineer, and a role word (engineer, developer, scientist, architect). Interns, managers, heads, sales, solutions, product, program, recruiting, marketing, design, legal, support and operations titles are removed. That leaves 4,995.
- Three largest posters set aside: the three employers with the most US postings overall are removed, which takes out 136 AI-titled postings and leaves 4,859.
- Skill named: the description matches a word-bounded pattern, for example
PyTorch,TensorFlow,RAGor retrieval-augmented,KubernetesorK8s,fine-tuning, orGoonly asGolangor in a language list. We checked ten random matches for the loosest ones (Go, Spark, Ray, evals, vector databases, agents) and tightened the patterns. - Seniority from the title: senior, sr, staff, principal, lead, head, director, distinguished or fellow is senior and up. Junior, entry, graduate or associate is junior. Everything else is mid or unspecified.
What to do with this
- Put Python and one LLM project at the top. They are the two most-named skills, and the LLM one is what separates today's postings from older lists.
- Name PyTorch, not just "deep learning". It is named about twice as often as TensorFlow.
- Show retrieval and evals work if you have it. RAG and evals each appear in about one posting in six, and mostly alongside LLMs.
- For senior roles, add the platform side: serving, Kubernetes, Spark, distributed systems.
- Match your resume wording to the posting. Applicant tracking systems filter on the exact tool names above, so use the posting's spelling.
FAQ
›What skills do AI engineer job postings ask for most?
Python (57.9% of 4,859 US postings), LLMs (48.4%), AI agents (40.4%), distributed systems (24.0%) and PyTorch (22.1%).
›PyTorch or TensorFlow: which is in more AI job postings?
PyTorch: 22.1% of postings against 11.4% for TensorFlow. JAX is named in 6.0%.
›Do AI engineer jobs require RAG and vector databases?
RAG is named in 18.3% of postings, vector databases in 9.2% and LangChain or LangGraph in 8.3%. Most RAG postings also name LLMs (88%).
›Do senior AI roles ask for different skills?
Mostly the same, with more platform words: Kubernetes 14.8% vs 11.5%, Spark 10.5% vs 7.4%, MLOps 16.8% vs 12.5% for senior and mid-level postings.
›Do I need Rust or Go for AI engineering?
Rarely. Go is named in 6.3% of postings and Rust in 2.8%, against 20.9% for C++ and 16.8% for Java.
Method and sources
- Source: OfferMagnet's own index of public postings. 77,814 US postings first seen 19 Sep to 30 Oct 2026; 4,995 have an AI or ML engineering title; 4,859 remain after setting aside the three employers with the most postings.
- Skills are matched in the first 4,000 characters of each description with case-aware, word-bounded patterns (for example Go only as Golang or inside a language list, Spark not in product names). Counts are a floor.
- Seniority comes from title words only. Pairs are counted over the 25 listed skills. Junior-titled postings (53) are too few for their own cells.
- We report counts of postings, not people. No applicant or customer data is used.