But when I look at the actual job listings — 2,340 AI-related roles across the startups we track — the reality is more nuanced and more interesting than the hype suggests. The AI job market isn't one market. It's at least four different markets with different entry points, different skill requirements, and wildly different compensation.
Let me break it down.
The four AI job markets
Market 1: AI Research (the PhD track)
These are the Research Scientist and Research Engineer roles at places like OpenAI, Anthropic, DeepMind, and a handful of well-funded AI labs. They pay extraordinarily well — $300K median for Research Scientists — but they're essentially inaccessible without a PhD in machine learning, computer science, or a related field, plus published papers at top conferences.
There are roughly 50-100 of these roles open at any given time. The competition is global and fierce. If you're not already on this track, this isn't your entry point.
Market 2: ML Engineering (the builder track)
Machine Learning Engineers build and maintain the models, training pipelines, and inference infrastructure that make AI products work. This is the largest segment of "AI jobs" by volume, and it's where most of the 2,340 roles in our data sit.
The median salary is $220K for ML Engineers, rising to $273K for Staff ML Engineers. The skills required are a blend of software engineering and ML: Python, PyTorch or TensorFlow, distributed systems, and enough math to understand what the models are doing (but not necessarily enough to invent new architectures).
This is the most accessible "real AI" role for experienced software engineers. If you have 3+ years of backend engineering experience and are willing to invest 6-12 months in learning ML fundamentals, this path is realistic.
Market 3: Applied AI Engineering (the integration track)
This is the fastest-growing segment and the one most people should be targeting. Applied AI Engineers take existing foundation models — GPT-4, Claude, Llama, Gemini — and make them useful in production applications. They build RAG pipelines, design prompt chains, implement evaluation frameworks, and figure out how to make LLMs reliable enough for real users.
The median salary is $250K, which is higher than general ML Engineering because the demand is outpacing supply. With 34 open "Applied AI Engineer" roles at top salary levels and hundreds more with similar job descriptions under different titles, this is where the market is hottest.
The skills are different from traditional ML: you need strong software engineering fundamentals, experience with LLM APIs and frameworks (LangChain, LlamaIndex, or equivalent), understanding of vector databases and retrieval systems, and — critically — the judgment to know when AI is the right solution and when it isn't.
Market 4: AI-Adjacent Roles (the non-technical track)
Not every AI job requires you to write code. AI companies need product managers who understand model capabilities and limitations. They need data analysts who can evaluate model performance. They need technical writers who can document APIs. They need sales engineers who can demo AI products to enterprise customers.
These roles don't show up when you search for "AI engineer," but they're real, they're growing, and they pay well. A Product Manager at an AI startup earns a median of $210K. A data analyst focused on model evaluation can earn $180K+.
Who's actually hiring
The AI hiring landscape is more concentrated than you might expect. Here are the companies with the most AI-related openings:
Waymo leads with 35 open AI roles — autonomous driving requires massive ML teams. OpenAI and Zoox each have 20. Pinterest (16), Reddit (15), and Roblox (14) represent the "AI integration" wave — established tech companies building AI features into existing products. Anduril (13) represents defense tech's growing AI appetite. Figure AI (10) and Scale AI (9) round out the top.
The pattern is clear: the biggest AI employers are either pure AI companies (OpenAI, Scale AI) or companies where AI is a core product differentiator (Waymo, Zoox, Anduril). The "we're adding AI to everything" companies (Pinterest, Reddit, Roblox) are a close second.
Where AI jobs are
San Francisco dominates with 316 AI roles — no surprise there. New York is second with 166, followed by London (115), Remote US (105), and Remote (96). Bangalore has 77, Mountain View has 68, and Boston has 34.
The geographic concentration is tighter than the overall startup job market. AI jobs are disproportionately in SF and the Bay Area, which makes sense given the density of AI companies and research labs in the region. If you want the widest selection of AI roles, being in or willing to relocate to the Bay Area gives you access to roughly 25% of all positions.
That said, 201 AI roles are fully remote (combining Remote US and Remote categories), which is about 9% of the total. Remote AI work exists, but it's less common than in other engineering disciplines.
The skills that actually matter
I looked at the tech requirements across all 2,340 AI roles. Here's what shows up most frequently:
Python is non-negotiable. It appears in virtually every AI job listing. If you're coming from a JavaScript or Java background, learning Python is step one.
PyTorch has overtaken TensorFlow as the dominant ML framework. If you're choosing one to learn, choose PyTorch.
Cloud platforms (AWS, GCP, Azure) appear frequently because AI workloads require serious compute infrastructure. Knowing how to provision GPU instances, manage training jobs, and deploy models on cloud infrastructure is increasingly expected.
LLM-specific skills — prompt engineering, RAG architecture, fine-tuning, evaluation — are the fastest-growing requirements. Two years ago, these skills didn't exist as job requirements. Now they appear in hundreds of listings.
SQL and data engineering skills (Spark, Kafka, Airflow) show up because most AI work starts with data. If you can't wrangle the training data, you can't build the model.
Kubernetes and Docker appear because deploying ML models in production requires containerization and orchestration. The gap between "I trained a model in a notebook" and "I deployed a model that serves 10,000 requests per second" is where these skills matter.
The realistic path in
Here's what I'd actually recommend based on the data, depending on where you're starting from:
If you're a software engineer with 2+ years of experience: Target Applied AI Engineering roles. You already have the software engineering fundamentals. Spend 3-6 months learning LLM APIs, RAG patterns, and vector databases. Build 2-3 projects that demonstrate you can take a model from prototype to production. The market for this profile is desperate — companies are hiring faster than candidates are entering.
If you're a data scientist or data analyst: Target ML Engineering roles. You already understand the math and the data. The gap is in software engineering practices: writing production-quality code, building CI/CD pipelines, and deploying models at scale. Bridge that gap and you're competitive for roles paying $220K+.
If you're early in your career (0-2 years): Don't try to go directly into AI. Get a solid software engineering or data engineering foundation first. The AI roles that hire junior candidates are rare and extremely competitive. The AI roles that hire people with 3-5 years of strong engineering experience are plentiful and well-compensated.
If you're non-technical: Target AI-adjacent roles at AI companies. Product management, technical program management, solutions engineering, and developer relations at AI startups are all growing functions. You need enough technical literacy to understand what the models do, but you don't need to build them yourself.
The compensation reality
AI jobs pay a premium. The median salary across all AI-related roles in our data is $250,000, with a 75th percentile of $300,000 and a 90th percentile of $350,000. That's roughly 20-40% above comparable non-AI engineering roles.
But the premium isn't evenly distributed. Research Scientists and Staff ML Engineers command the highest salaries. Applied AI Engineers are close behind. Junior or mid-level roles in AI pay well but not dramatically more than equivalent non-AI positions.
The premium also has a shelf life. As more engineers upskill into AI, the supply-demand imbalance will narrow. The people who enter the field now will benefit from the current premium. The people who enter in 2-3 years will face more competition and potentially lower premiums.
The honest assessment
Breaking into AI in 2026 is realistic but not easy. The roles are real, the salaries are exceptional, and the demand is genuine. But the bar is high — companies want people who can ship AI products, not people who completed an online course and added "AI/ML" to their LinkedIn headline.
The 2,340 AI roles in our database represent about 2% of all startup jobs. It's a meaningful market, but it's not the entire market. If AI isn't the right fit for your skills or interests, there are 102,535 other startup jobs that need talented people.
And if AI is the right fit — if you're genuinely excited about the technology and willing to invest the time to build real skills — the data says the market will reward you. $250K median, growing demand, and a field that's still early enough that the best opportunities haven't been claimed yet.
The window is open. It won't stay open forever.
