A curated guide — not a job board

Where to find AI engineering jobs (and how to read the postings)

Where to look

• Vetted marketplaces (e.g. arc.dev) — curated, interview-ready mid/senior roles. You pass their own technical and communication screen first; a portfolio and live performance matter more than a cert here.
• Startup boards — Wellfound (AngelList Talent) and Y Combinator's Work at a Startup are where "first AI hire / builder-architect" roles cluster.
• Cloud and AI partners — AWS Partner Network and the Anthropic / Claude Partner Network member directories list consultancies hiring GenAI engineers and architects; many value vendor certs alongside a track record.
• Company career pages directly — for specific labs and scale-ups, the freshest agentic roles often appear on their own sites before the aggregators.
• Community job channels — practitioner newsletters and communities (for example Latent Space's audience) surface roles that never hit the big boards.
• LinkedIn — still the widest net; filter for "agentic," "LLM," "GenAI," and the specific platforms (Bedrock, Claude, Cloudflare) rather than just "AI."

How to read an AI job posting

Decode the title

"Agentic AI Engineer," "AI Agent Architect," "GenAI Solutions Architect," and "AI Engineer" are largely interchangeable in 2026. Read the responsibilities, not the title, to tell build-heavy from design-heavy from client-facing. (See the roles, decoded.)

Tell "build with models" from "build models"

If the requirements emphasize the agentic loop, tool use, RAG, evals, prompt engineering, and deployment, it's an AI-engineering role you can target without a machine-learning background. If they emphasize training models, PyTorch, MLOps pipelines, and ML theory, that's an ML role — a different job. (See the distinction.)

Separate must-haves from wish-lists

AI postings are notorious for kitchen-sink requirements (every framework, every cloud, sometimes years of experience in tools barely that old). Treat the list as a wish list: the real bar is usually a handful of core skills plus evidence you can ship. Apply if you hit the core, even with gaps.

Spot the green and red flags

• Green: mentions of evals, observability, cost control, safety/guardrails, and shipping to production — signs of a serious, mature team.
• Red: "prompt engineer" as the whole job, no mention of evaluation or production concerns, or impossible "5+ years with [tool released last year]."