4 New AI Job Roles That Didn't Exist 18 Months Ago
Vibe Coding Engineer. AI Safety Classifier Engineer. Agent Orchestration Specialist. MCP Integration Developer. Here's what each role actually does, what it pays, and how to position for it.
The Job Titles Nobody Was Searching For in 2025
In January 2025, you couldn't find a single job posting for a "Vibe Coding Engineer." The title didn't exist. Today, LLMHire surfaces hundreds of postings across Greenhouse, Lever, and Ashby that use the phrase — or close variants like "AI-Native Developer," "Agentic Software Engineer," or "LLM Application Engineer" — and hiring managers increasingly expect what those titles imply: fluency with Claude Code or Cursor as primary development environment, not just a productivity add-on.
The AI coding tooling revolution created job categories faster than universities could create curricula for them. Here are four roles that materialized in 2025-2026, what they actually require, and where they're hiring.
1. Vibe Coding Engineer
What it is: A software developer whose primary development environment is an AI coding agent — Claude Code, Cursor, or similar — and who is explicitly hired for their ability to ship AI-assisted products fast. The term emerged from the Andrej Karpathy "vibe coding" concept: telling the AI your intent and navigating to the result rather than writing every line.
What distinguishes it from "just using Copilot": Vibe Coding Engineers are expected to manage agentic workflows across multi-file projects, write effective CLAUDE.md files and system prompts, understand where to intervene (security, architecture, data modeling) versus delegate, and ship faster than a traditional engineer.
Where it's hiring: Early-stage startups (seed to Series B) building on AI-native stacks. Product-led growth companies that need to move faster than enterprise development teams. Notable concentration in fintech, healthcare AI, and developer tools.
Salary range (July 2026): $120K–$185K at startups; $160K–$240K at Series B+ with equity.
How to position for it: Build a public portfolio of AI-assisted projects. Document your CLAUDE.md configurations. Browse Vibe Coding Engineer roles →
2. AI Safety Classifier Engineer
What it is: An ML engineer who builds, trains, and red-teams the safety classifiers layered over large language models — the systems that caught the Fable 5 exploit in June 2026 and enabled its return after a 19-day suspension. This role sits at the intersection of ML engineering and adversarial security research.
What the work looks like: Writing adversarial test cases designed to elicit harmful outputs. Training binary classifiers on curated datasets of safe/unsafe model outputs. Measuring false positive rates against legitimate use cases. Maintaining classifier performance as the underlying model evolves.
Why it's exploding: The Fable 5 export-control incident made safety classifiers a regulatory requirement, not just a best practice. Every frontier lab, and increasingly every enterprise deploying frontier models, needs this capability in-house. Anthropic, OpenAI, Google DeepMind, Meta, and Cohere are all hiring for variants of this role.
Salary range: $170K–$280K at frontier labs; $140K–$220K at enterprises deploying models internally.
How to position for it: Red team experience, adversarial ML research, experience with Constitutional AI or RLHF. Browse AI Safety roles →
3. Agent Orchestration Specialist
What it is: A developer who designs, builds, and debugs multi-agent systems — networks of AI agents that coordinate to complete complex tasks that no single agent can handle alone. The role is part architect, part debugger, part prompt engineer.
Looking for AI-native engineers?
Post your role for free on LLMHire and reach thousands of verified engineers actively exploring opportunities.
What the work looks like: Designing task decomposition strategies (which sub-tasks get which agent). Writing orchestration code in frameworks like Claude Agent SDK, LangGraph, or Temporal. Building observability and retry logic for agents that fail mid-task. Optimizing token spend across multi-agent chains where costs compound.
Why it's distinct from "AI engineer": Most AI engineers integrate a single LLM API call into an application. Agent orchestration specialists design the coordination layer — what happens when Agent B gets a result from Agent A that contradicts what Agent C expected, while Agent D is still running.
Where it's hiring: Enterprise automation teams, AI-native startups building autonomous workflows, financial services firms automating research and compliance pipelines.
Salary range: $150K–$220K; $200K–$300K+ at hedge funds and financial services.
How to position for it: Build a multi-agent project publicly. Demonstrate observability and failure handling, not just happy-path orchestration. Browse Agent Orchestration roles →
4. MCP Integration Developer
What it is: A developer who builds and maintains Model Context Protocol (MCP) servers — the integration layer that gives AI coding tools like Claude Code access to external systems (databases, APIs, internal tools, enterprise software). As Claude Code's MCP ecosystem exploded in 2025-2026, companies discovered they needed specialists to own this infrastructure layer.
What the work looks like: Building MCP servers that expose company-specific tools to AI agents. Writing TypeScript or Python MCP server implementations. Maintaining security for MCP authentication and permissions. Debugging AI tool behavior caused by MCP data format or response issues.
Why enterprises care: A well-built MCP server lets Claude Code access a company's internal APIs, Jira, Confluence, databases, and custom tooling — dramatically multiplying what developers can do with AI coding tools. A poorly-built one creates the security exposure documented in the June 2026 MCP config-file-trust CVE cluster.
Salary range: $130K–$190K; this role often pays a 15-25% premium over equivalent backend developer roles due to scarcity.
How to position for it: Build a public MCP server and document it. Contributions to the official MCP specification or popular open-source MCP server packages carry significant weight. Browse MCP Developer roles →
What Ties These Roles Together
All four roles emerged from the same structural shift: AI went from a tool developers use to an agent that developers orchestrate. The jobs aren't about using AI to code faster — they're about designing, securing, and coordinating AI systems that do substantial work autonomously.
The common thread in hiring criteria across all four: demonstrated real-world project work with AI coding tools, not certifications. Hiring managers at the companies posting these roles are explicitly looking for people who have shipped something using the tools, not people who completed a course about them.
If you're building toward any of these roles, the Vibe Coding Ebook covers the technical landscape for vibe coding engineering in depth. Vibe Coding Academy has structured learning paths for Claude Code and agent orchestration. EndOfCoding tracks how these job categories are evolving week-to-week.
LLMHire tracks 5,954+ AI engineering roles from Greenhouse, Lever, Ashby, and direct company listings. Updated 6× daily.
Browse all AI roles → · AI Safety roles → · Agent Orchestration → · MCP Developer →