A new job category is forming around AI reasoning systems.

These roles didn't exist 18 months ago. By 2028, they'll be mainstream.

What reasoning engineers build: systems where AI doesn't just generate — it thinks. Multi-step problem solving. Self-verification loops. Chain-of-thought architectures. Systems that know when they're uncertain and seek more information.

Specific roles: Reasoning System Architect (designing how AI breaks down complex problems). Verification Engineer (building systems that check AI's own work). Planning System ML Engineer (AI that creates and executes multi-step plans). Inference Optimization Engineer (making reasoning fast enough for production).

Why this is exploding: reasoning capabilities are what separate AI that assists from AI that acts. An agent that can genuinely reason through a tax scenario, a legal question, or an engineering design problem is orders of magnitude more valuable than one that just generates text.

Companies investing heavily: Anthropic, OpenAI, Google DeepMind, and every company building serious agent systems.

Salary: $180K-$350K. The reasoning layer is considered the highest-value component of modern AI systems.

Skills needed: deep understanding of transformer architectures, reinforcement learning (reasoning models often use RL), evaluation methodology (how do you measure reasoning quality?), and formal logic.

This is the frontier of AI capability. The engineers working on it are building the systems that will define what AI can do in the next decade.