Hiring AI engineers in 2026: the brief most firms still get wrong

Applied AI Engineering — the work of building, evaluating, and shipping AI-powered features into production — is now a real engineering discipline. There are real best practices, real failure modes, and real people who’ve done the work for long enough to be experts.

The hiring briefs we’re seeing, though, still mostly describe a generic Senior Engineer with the word “AI” bolted onto the job title. This costs firms time, money, and the candidates they actually want.

What Applied AI Engineering actually is

The work spans four areas, in roughly declining order of how often it shows up in real product engineering: building evaluation infrastructure for AI features, designing and operating RAG systems, the careful engineering around prompt construction and model selection, and the infrastructure around inference (serving, caching, fallbacks).

Notice what isn’t on that list: training models from scratch. The overwhelming majority of applied AI engineering at scale-ups in 2026 is not training models. It is shipping product features that use models — most often third-party — in a way that is reliable, evaluable, and cost-effective.

What the briefs still say

“Senior AI Engineer — experience with LLMs, RAG, and fine-tuning required. Familiarity with PyTorch and Hugging Face transformers. ML Ops experience a plus.”

This brief reads as written by someone whose mental model of AI engineering is from 2023. Fine-tuning is now rare at scale-ups (foundation models have gotten good enough). PyTorch knowledge matters at maybe 5% of the roles we work on. ML Ops as a phrase is increasingly archaic.

What the brief should say:

“Senior Applied AI Engineer. We have an AI-backed feature in production that handles X queries a day. We’re looking for the engineer who can take ownership of the evaluation infrastructure, the RAG retrieval quality, and the cost/latency optimisation. Comfortable with prompt design and model selection. Healthy attitude towards using third-party models where appropriate.”

That’s a real brief. It will close real candidates.

What we tell clients

If your AI engineering brief mentions training, fine-tuning, or “researching new model architectures” as primary responsibilities — and the role is not at a foundation model lab — the brief is probably mis-specified. The conversations with strong applied AI engineers will go nowhere because the candidates will spot the mismatch immediately.

We’re happy to walk through the brief before you publish it. Most of these conversations are 30 minutes and save the firm two months of search.