Find Your Frontier — Where to Plant Your Flag
AI Is Five Careers, Not One
The phrase "I want to work in AI" hides five very different careers behind one label. NLP engineers and robotics engineers do almost nothing in common. Multimodal product builders and frontier research scientists live in different universes. Picking the wrong field for your strengths means spending five years swimming upstream. Picking the right one means your existing experience becomes a shortcut. The point of this lesson is to give you a map so you can choose deliberately, instead of chasing whatever was loudest on social media this month.
- NLP & Language: language models, search, conversational systems — the broadest entry door in 2026
- Computer Vision: perception, imaging, autonomous systems — physical-world AI with deep specialist career paths
- Multimodal: products that combine text, image, audio, video — where most consumer AI ships today
- Robotics & Embodied AI: AI that acts in the physical world — long horizons, high leverage, hardware-bound
- Frontier research: pushing the underlying model capability itself — narrow door, very long horizons, very large impact
Where the Jobs Are by Field
Each frontier has a distinct shape: who the typical employers are, what the role categories look like, and what a credible portfolio looks like. The fastest way to orient yourself is to look at job listings from three or four representative employers in the field you're considering. The role titles tell you what the field actually values.
- NLP roles cluster around: prompt engineer · LLM app developer · NLP scientist · conversational designer · search/relevance engineer
- Computer Vision roles cluster around: CV engineer · perception scientist · ML ops for edge devices · imaging specialist · annotation pipeline lead
- Multimodal roles cluster around: multimodal engineer · applied AI engineer · agent builder · voice/video product engineer
- Robotics roles cluster around: robotics engineer · controls or RL scientist · simulation engineer · hardware-ML hybrid · safety case engineer
- Frontier research roles cluster around: research scientist · pre-training engineer · alignment researcher · evals researcher — almost always PhD or strong open record
How to Pick Yours — Three Honest Heuristics
Most career advice in AI is bad because it optimises for the giver, not the receiver. Three heuristics actually help. First: where does your existing background give you a shortcut? Second: what kind of problem do you actually enjoy thinking about for ten hours straight? Third: what are you willing to commit to for five years, given that AI fields compound on accumulated context? You don't have to get it perfectly right — but you do have to choose deliberately, because momentum in AI comes from going deep, not from hopping.
- Existing leverage: a backend engineer has a shortcut into NLP/LLM apps; a mechanical engineer has a shortcut into robotics; a designer has a shortcut into multimodal product
- Problem temperament: do you enjoy debugging language, debugging perception, or debugging physics? Each field is mostly that, every day
- Time horizon: NLP/multimodal pays off in months; robotics and frontier research pay off in years — match the horizon to your patience and runway
- Adjacent paths: AI product manager, AI consultant, security SE, safety & policy, data engineer, designer for AI surfaces — high-leverage roles that don't require shipping models
- Don't chase the hottest field — pick the one where your existing strengths give you the largest unfair advantage
Prefer slides, quizzes, and saved progress? Read this lesson in the library — free, no sign-up.