The most reliable way to fill AI roles internally is to move strong adjacent performers, backend engineers into AI engineering, QA into eval engineering, analysts into AI data work, through a structured path: a real project, a named mentor, and protected time, not a course subscription. Internal movers already know your domain and systems, which is half the job; the course-and-hope approach fails because it skips the project.
The transitions that work
| From | To | What transfers | What's new |
|---|---|---|---|
| Backend engineer | AI engineer | Systems, APIs, production discipline | LLM behavior, prompting, RAG, evals |
| QA engineer | AI QA / eval engineer | Test design, edge-case instinct, rigor | Statistical evaluation, LLM failure modes |
| Data analyst | AI data engineer / annotator lead | Data fluency, quality instinct, SQL | Pipelines, labeling ops, embeddings |
| Domain expert | AI trainer / context designer | Ground truth, judgment, taxonomy | Prompting, rubric design, eval thinking |
| Frontend engineer | AI product engineer | UX instinct, product proximity | Streaming UX, model integration, failure handling |
The program that makes it stick
- 1Select for pull, not push: take volunteers who already tinker with AI tools, not people you need to relocate off a sunsetting project.
- 2Anchor on a real deliverable: a production AI feature or eval suite, scoped to a quarter.
- 3Assign a named mentor, an in-house senior or an embedded AI engineer, with weekly pairing.
- 4Protect 50-100% of their time; 'learn AI on Fridays' programs quietly die.
- 5Make it official at the end: new title, new scorecard, adjusted pay, or the person and the lesson both walk.
Where mobility beats hiring, and where it doesn't
- Mobility wins when domain context dominates: your data, your customers, your systems.
- Hiring wins for the first senior AI hire, someone must set the bar the internal movers learn against.
- The blend: one senior external anchor per pod, internal movers around them, is the pattern that compounds.
- Never staff a whole AI team by mobility alone; without an experienced anchor, everyone relearns public knowledge slowly.
