How to find an AI specialist.

The prompt engineer title is being absorbed into broader roles faster than almost any job title in recent memory, because prompting turned out to be a component of building AI systems rather than a discipline in itself. Many people write a good prompt. Far fewer can tell you whether a change improved anything.

That is where the actual scarcity sits: systematic evaluation, retrieval that grounds answers in real data, and understanding failure modes in production. The field also attracts an unusually high volume of shallow claims, which makes public artefacts more valuable here than in almost any other role.

Job titles worth searching

Grouped by what the person actually does, because searching all47 at once produces a result set you cannot triage. Decide which group you need first — that decision does more for the search than any string below.

Where AI practitioners actually are

Published work separates building from describing, which matters more in this field than any other because claims so far outpace capability. Repositories containing retrieval pipelines, agent orchestration, or evaluation harnesses demonstrate real work; a compelling description of AI experience demonstrates nothing.

Honest writing about failure is an unusually strong filter. Anyone can describe a successful demo, but explaining why a chunking strategy failed or why an evaluation set gave misleading results reveals someone who has operated a system in production rather than built one for a presentation.

Two pools are consistently overlooked. Conversation designers from linguistics and UX writing backgrounds are frequently better at prompt precision than engineers, since language exactness is their existing craft. And domain experts who became AI-fluent — lawyers, clinicians, analysts — can recognise when an output is confidently and subtly wrong, which is the judgement that determines whether an application is safe to deploy.

Boolean search strings

Written to be pasted as-is. Each one is built around an intent rather than a platform, since the useful question is what you are trying to find, not which site you happen to be on.

site:linkedin.com/in/ ("AI engineer" OR "prompt engineer" OR "applied AI") ("RAG" OR "evaluation" OR "fine-tuning" OR "LLM") "{city}"

site:github.com ("langchain" OR "llamaindex" OR "dspy" OR "instructor" OR "eval") ("retrieval" OR "agent" OR "pipeline") -tutorial -awesome

("eval set" OR "evaluation harness" OR "LLM as judge" OR "regression testing" OR "golden dataset") ("LLM" OR "model") -course -vendor

("RAG" OR "retrieval augmented" OR "chunking strategy" OR "embedding model" OR "reranking") ("production" OR "latency" OR "accuracy") -course -vendor

("lawyer" OR "clinician" OR "analyst" OR "researcher") ("using AI" OR "LLM" OR "built a tool" OR "automated") ("in my practice" OR "in our team") -jobs -course

(site:substack.com OR site:medium.com OR site:github.io) ("we built" OR "lessons from" OR "what failed") ("LLM" OR "AI agent" OR "RAG") -course

("conversation design" OR "conversational AI" OR "dialogue design") ("chatbot" OR "voice" OR "assistant") -jobs -vendor -course

("AI hackathon" OR "LLM hackathon" OR "prompt competition" OR "AI engineer summit") ("winner" OR "speaker" OR "built") 2024..2026 -jobs

Mistakes that cost the most time

1. Hiring for prompting when the job needs evaluation

   Writing a good prompt is a skill that many people can develop quickly. Building systematic evaluation — test sets, quality criteria, regression detection, and knowing whether a change improved anything — is genuinely difficult and rare. Most production AI problems are evaluation problems, and hiring for prompting alone leaves them unsolved.

2. Underestimating the volume of shallow claims

   This title attracts an unusually high proportion of candidates whose experience is using consumer AI tools competently. That is not nothing, but it is not building production systems. Public artefacts — repositories, published systems, honest writing about failures — separate the two quickly where interviews frequently do not.

3. Overlooking domain experts who became AI-fluent

   A lawyer or clinician who can evaluate AI output knows when an answer is confidently and subtly wrong, which a generalist AI specialist often cannot detect. For domain-specific applications this judgement matters more than technical AI depth, and these candidates are far less contested.

4. Treating the title as stable

   Prompt engineering as a standalone job is already being absorbed into broader roles, because prompting turned out to be one component of building AI systems rather than the whole discipline. Hiring against the title rather than the actual work risks a role that looks outdated within a year.

5. Ignoring conversation designers and linguists

   People from linguistics, technical writing, and UX writing backgrounds are frequently better at constructing precise prompts than engineers, because language precision is their existing craft. They are rarely considered for AI roles, which keeps them available.

6. Screening on framework familiarity

   The AI tooling landscape changes faster than any other in technology, and today's framework may be legacy within a year. What transfers is understanding the underlying problems — grounding, evaluation, latency, cost, failure modes — since frameworks are learned in days by anyone who grasps those.

Common questions

What job titles should I search for when hiring an AI specialist?

Search the work rather than the title, since this vocabulary is shifting rapidly. For system building, RAG Engineer, AI Agent Engineer, AI Integration Engineer, and Applied AI Engineer. For the maturing evaluation discipline, AI Evaluation Engineer and Model Evaluation Specialist. For language-focused work, Conversation Designer and AI Content Designer, roles often filled from linguistics and UX writing. Domain-embedded titles such as Legal AI Specialist and Clinical AI Specialist identify the most valuable and least contested group.

Is prompt engineering still a real job?

As a standalone title it is already being absorbed into broader roles, because prompting proved to be one component of building AI systems rather than a discipline in itself. What remains genuinely valuable and scarce is everything around it: designing evaluation that detects whether a change actually improved outputs, building retrieval systems that ground responses in real data, orchestrating multi-step agent behaviour, and managing cost and latency in production. Hiring for prompting alone typically leaves the harder problems unaddressed.

How do I tell genuine AI capability from confident claims?

Look at artefacts rather than descriptions, because this field attracts a high volume of shallow claims. Published repositories showing retrieval pipelines, agent systems, or evaluation harnesses demonstrate building rather than describing. Writing about what failed is particularly informative — anyone can describe a successful demo, but explaining why a retrieval strategy did not work reveals genuine experience. In interview, questions about evaluation methodology and failure modes separate practitioners quickly, because those are the problems that only appear in production.

Should an AI specialist come from a technical or domain background?

It depends on the application, and domain background is more often the right answer than recruiters assume. For building AI infrastructure and pipelines, engineering ability is essential. But for applying AI within a specific field — legal, clinical, financial — a domain expert who became AI-fluent can evaluate outputs in ways a generalist cannot, because they recognise when an answer is confidently and subtly wrong. That judgement is frequently the binding constraint on whether an AI application is safe to deploy.

Where can I find AI practitioners outside LinkedIn?

GitHub is the strongest source, since published retrieval systems, agent frameworks, and evaluation harnesses demonstrate real building. Technical writing that discusses failures rather than successes identifies people who have shipped something. The applied AI community runs frequent hackathons and events whose participants are demonstrably building. Two under-searched pools are worth attention: conversation designers from linguistics backgrounds, who are often better at prompt precision than engineers, and domain experts who became AI-fluent within their own field.