Skip to main content

Role · AI Hiring Index · as of 2026-11-19

AI Engineer jobs at AI companies

Sierra (29) and Wayve (20) lead 183 open AI Engineer roles across 39 AI companies. 10% are remote and 51% are in the Bay Area. The November 2026 Index counted 5 fewer openings than October 2026, among companies measured in both months.

AI engineers build the product features that run on language models and the systems underneath them — retrieval, agents, evaluation, and fine-tuning where prompting is not enough. At model companies many also train and serve models, which is why PyTorch appears in so many of these postings.

Product engineering at AI-native companies; applied research or platform teams at the labs.

Demand this month

Open roles
183
−5
New in the last 7 days
8
Remote
10%
Bay Area
51%

What AI Engineer postings ask for

Share of 183 open postings that mention each skill; change since the previous Index in points.

SkillShare, as a barShare
Evals61%n=112+7 pt
LLMs55%n=100−12 pt
Python51%n=94−4 pt
Agents46%n=84−7 pt
RAG / retrieval37%n=67+1 pt
Go30%n=55+4 pt
Prompting26%n=47±0 pt
PyTorch25%n=46−2 pt
TypeScript23%n=42−3 pt
C++19%n=34+8 pt
Fine-tuning18%n=33+2 pt
RL / post-training18%n=33−3 pt
Kubernetes15%n=28+3 pt
Distributed training9%n=17±0 pt
Rust9%n=16+5 pt
PhD8%n=14+10 pt
CUDA / Triton7%n=12±0 pt
vLLM / SGLang / TensorRT5%n=10±0 pt
MCP5%n=9+3 pt
JAX4%n=8±0 pt

Posted base pay

Median posted range $196K–$324K base, from 104 postings that publish pay (57% of this role's openings). This is the base range a posting advertises — not total compensation, which adds equity and bonus; for that, Levels.fyi is the reference.

Midpoint of each posted base range, in thousands of US dollars: <150: 0; 150–200: 14; 200–250: 30; 250–300: 27; 300–350: 22; 350–400: 5; 400–450: 3; 450+: 3

Half of the posted ranges have their midpoint between $220K and $310K; the median midpoint is $264K.

What to learn

What you can skip

Backend and API design, data modelling, testing, deployment and on-call — the software engineering these roles are built on. What changes is that model behaviour is now part of the system you test.

AI Engineer

After the foundations: retrieval, agents and tools in a product, measured; then changing the model itself when prompting stops being enough, and serving it at a cost the product can carry.

  1. Stage 1Build LLM features

    Ship retrieval and tool use in a product, with numbers behind each.

    1. Agents · 46% of AI / ML Engineer postings (n=183) · about 1–2 weeks

      Write a tool-using agent loop yourself and measure where it fails.

      • Building effective agents — Anthropic free

        article · 15 minutes — Workflows versus agents, and when you need neither: the handful of patterns these systems are built from, named by a lab that ships them.

      • Hugging Face Agents Course free

        course · about 15–20 hours — Hands-on: tool calling, a ReAct-style loop and multi-agent setups in code you run — the gap between reading about agents and writing one.

      • Effective context engineering for AI agents — Anthropic free

        article · 20 minutes — Context is the agent loop's most expensive variable: when to trim, compress, take notes or hand off to a sub-agent, sorted by problem — the checklist for your own loop's context budget.

      Checkpoint: Write the agent loop yourself, then evaluate it — Messages API tool use with a step cap, tool-error retries and context trimming; score it on 30 real tasks for tool choice, argument correctness and steps taken.

    2. RAG / retrieval · 37% of AI / ML Engineer postings (n=183) · about 1 week

      Retrieve the right context for a question and measure how often you do.

      • Introducing Contextual Retrieval — Anthropic free

        article · 15 minutes — Hybrid search (embeddings plus BM25), reranking and chunk context, each with a measured drop in retrieval failures — design with numbers.

      Checkpoint: Hybrid retrieval with a labelled query set — BM25 plus vectors fused with reciprocal rank fusion, a reranker, and 50 hand-labelled queries; report recall@k and nDCG against keyword search alone.

    3. MCP · 5% of AI / ML Engineer postings (n=183) · about 3 days

      Expose a real system to agents through MCP, and measure whether they use it correctly.

      • Model Context Protocol — introduction and spec free

        docs · 45 minutes — The protocol itself — tools, resources, transports — straight from the spec rather than from a framework's wrapper around it.

      Checkpoint: An MCP server for an API you know, measured by an agent — Expose five real operations as tools, then run an agent on 30 tasks and count how often it picks the right tool with the right arguments; rewrite descriptions until that number moves.

  2. Stage 2Measure and improve

    Know when a change made things worse, and improve past what a prompt can do.

    1. Evals · 61% of AI / ML Engineer postings (n=183) · about 1 week

      Turn error analysis on production traces into evals that gate every change.

      • AI Evals: Everything You Need to Know — Hamel Husain and Shreya Shankar free

        article · 1.5 hours — Start from error analysis on real traces, not from a metric; then judges you check against people — the practice behind 'evaluation frameworks' in these postings.

      • promptfoo free

        repo · an afternoon — Runs an eval set as CI — cases in YAML, assertions by rule or by judge, red when the score drops — turning one-off eval studies into a regression suite. Open source; part of OpenAI since March 2026.

      • Inspect free

        repo · half a day — An evaluation institute's open-source framework: a clean task / dataset / solver / scorer split, with agent tasks and tool use first-class — the shape to grow a reusable harness in.

      Checkpoint: An eval harness that gates a change — Label 100 real outputs by hand, build an LLM judge and measure its agreement with you, and make CI fail when a prompt or model change drops the score.

    2. Fine-tuning · 18% of AI / ML Engineer postings (n=183) · about 2 weeks

      Fine-tune a small open model on one narrow task and prove it beats a prompted baseline.

      • Hugging Face LLM Course free

        course · 3 chapters, about 20 hours — Tokenizers, the Trainer, and fine-tuning chapters in runnable notebooks — enough to fine-tune a small open model on your own data.

      • TRL documentation — Hugging Face free

        docs · 1–2 hours — The library behind most supervised fine-tuning and preference tuning (DPO, GRPO) you will be asked about, with working recipes.

      Checkpoint: Fine-tune against a prompted baseline — LoRA-tune a small open model on one narrow task and compare it with a prompted frontier model on the same eval set — quality, latency and cost per 1,000 calls.

  3. Stage 3Train and post-train

    Read, change and train model code, not only call it.

    1. PyTorch · 25% of AI / ML Engineer postings (n=183) · about 3 weeks

      Train a small language model and change one design choice with an ablation to show for it.

      • Neural Networks: Zero to Hero — Andrej Karpathy free

        course · about 15 hours + coding — Backprop to a GPT in PyTorch, one video at a time — the fastest way for an engineer to read and debug model code.

      • Learn the Basics — PyTorch tutorials free

        docs · 2 hours — Tensors, autograd, datasets and the training loop in the framework's own words; a reference you will return to.

      Checkpoint: Train a small GPT and change one thing — Reproduce a small model with nanochat or nanoGPT on a modest dataset, change one design choice, and show the ablation with its loss curves.

    2. RL / post-training · 18% of AI / ML Engineer postings (n=183) · about 2 weeks

      Post-train a small model with preference or verifiable rewards, and see where it games the reward.

      • RLHF book — Nathan Lambert free

        book · about 12 hours — Post-training as practised on language models — reward models, PPO, DPO, verifiable rewards — by a researcher who post-trains open models, free to read online.

      • TRL documentation — Hugging Face free

        docs · an afternoon — Runnable GRPO and DPO trainers, so the book's methods become an experiment on a laptop-sized model.

      Checkpoint: Preference- or reward-tune a small model on a verifiable task — GRPO on arithmetic or unit-test-checked code with a small open model; plot reward against a held-out eval and explain where it starts to game the reward.

  4. Stage 4Serve it

    Run a model with the latency and cost a product needs.

    1. vLLM / SGLang / TensorRT · 5% of AI / ML Engineer postings (n=183) · about 1 week

      Serve an open model and measure latency and throughput under load.

      • LLM Inference Handbook — Modular free

        docs · 3–4 hours — Latency, throughput and cost of serving a model — batching, KV cache, quantization, which engine when — as one practical handbook rather than twenty blog posts.

      • Inside vLLM: Anatomy of a High-Throughput LLM Inference System free

        article · 1.5 hours — One request followed through the scheduler, paged KV cache, continuous batching, prefix caching and speculative decoding, ending on the latency-against-throughput curve a design answer needs.

      • vLLM documentation free

        docs · half a day — The engine the cloud-GPU route runs on: the docs explain the trade-off behind each setting, and the bundled vllm bench serve prints TTFT, TPOT and throughput — the deployment project's numbers.

      Checkpoint: Serve a small open model and measure it — Serve a small open model with vLLM, then measure time to first token and tokens per second at 1, 8 and 32 concurrent requests, with and without quantization.

One project

An LLM feature with its own eval harness — Build one feature on retrieval or tool use, label a test set by hand, gate changes on it in CI, and publish the error analysis — the portfolio piece hiring managers ask you to walk through.

The AI Engineer track on the learning map

Open AI Engineer roles

See all 185 postings

Posted demand, not hires: open postings on the public job boards of the AI companies we track, read every night. A repost counts once; a role family is assigned from the title; skill tags are checked against the postings every month and a skill below 90% precision is not shown; pay is the posted base range, shown only when at least 10 postings publish it. How we count.

Spot a mistake on this page?