The title peaked and fell, the skill did not
Interest in the Prompt Engineer title rose fast and fell almost as fast. Fortune reported in May 2025 that Indeed searches for the role jumped from two per million US searches in January 2023 to 144 per million in April 2023, then settled at around 20 to 30 per million. The same report cites Microsoft’s 2025 Work Trend Index, a survey of 31,000 workers in 31 countries, in which Prompt Engineer ranked second to last among the new roles companies were considering.
The work did not disappear. It stopped being a job on its own. Writing clear instructions for a model became something every engineer who works with AI is expected to do, the way writing a query became part of every backend role.
What AI-native engineering describes
Writing a good prompt turned out to be the easy part. The hard part is everything around it: what gets retrieved, how it’s chunked, what stays in memory between turns, which tools the model can reach, and how you know when any of it degrades.
AI-native engineering is the term gaining ground for engineers who treat model behaviour as a system to be built, tested and monitored. An AI-native engineer reaches for a model where it helps, measures what it does and treats its failures like any other bug. It’s a more useful description of the skill than any job title that came before it.
Context engineering: the work around the prompt
Most of the quality of an AI feature comes from the context the model receives, which is why the work is now often called context engineering.
- Retrieval. Finding the right documents or records for each request, and leaving out the ones that confuse the model.
- Chunking and indexing. Splitting content so that each piece makes sense on its own and can be found again.
- Memory. Deciding what the model remembers between turns, and what it should forget.
- Tools. Giving the model the functions it may call, with clear limits on what each one can do.
- Evaluation. A set of real examples with expected results, run every time anything changes, so quality is measured instead of guessed.
It looks more like building a search system than crafting a sentence, and it needs the same engineering habits: version control, tests, monitoring and code review.
The tools an AI-native engineer works with
The toolset changes every quarter, but the categories are stable:
- Coding assistants such as GitHub Copilot, Cursor and Claude Code, used every day and reviewed like any other contribution.
- Orchestration frameworks such as LangGraph and LlamaIndex, or workflow platforms such as n8n, to connect models, tools and data.
- Vector stores and search such as pgvector, Pinecone or Elasticsearch, for retrieval.
- Evaluation and tracing tools such as LangSmith or Langfuse, to see what a model received, what it returned and how that changed between versions.
- Model APIs and open-weight models, from OpenAI, Anthropic and Google to Llama and Mistral running on your own infrastructure.
An AI-native engineer knows which category a problem belongs to before picking a tool.
What an AI-native engineer does day to day
A typical week for an AI-native engineer on a product team includes:
- adding a model-backed feature behind a flag, with an evaluation set written before the feature ships
- tracing a bad answer back to the retrieval step that caused it
- comparing two models on cost, speed and quality for the same task
- setting token budgets and alerts so a runaway loop doesn’t become an invoice
- reviewing a colleague’s agent workflow the way they’d review any pull request
None of that is prompt writing in isolation. All of it uses prompting as one tool among many.
How to test for AI-native engineering when you hire
Interviews that ask candidates to write a clever prompt test the easy part. Better tasks look like real work:
- Fix a retrieval step that returns the wrong documents, and explain how you’d prove the fix works.
- Write an evaluation set for a feature, and say which failures it would catch and which it would miss.
- Plan for a model change. The provider retires the model next quarter. What do you check before switching?
- Talk about cost. How would you notice spend creeping up, and what would you do about it?
Strong candidates talk about measurement, failure cases and trade-offs. Weaker ones talk about wording.
AI-native engineering across every discipline
The shift reaches beyond AI specialists. QA engineers use models to generate test cases and then check them. DevOps engineers use AI assistants for infrastructure code and incident analysis. Frontend and backend developers use coding assistants every day, and the good ones review what the assistant writes as carefully as code from a junior colleague. In each case the skill is the same: knowing when a model helps, and verifying what it produces.
What AI-native engineering means for your team
Hiring for prompt skill alone buys one part of a role. What you need is engineers who can build and run the system around the model, and who test model behaviour as carefully as any other code. This is the profile companies look for when they hire AI engineers as part of a dedicated team.
For the same reason, every Azendo specialist in Chiang Mai and Bangkok completes AI certification through our Talent Success and Academy programme before joining a client team, whatever their discipline, and keeps developing through monthly coaching. With ordinary offshore staffing or staff augmentation, keeping an engineer’s AI skills current is your problem. In a fully managed team it’s ours, and it’s part of the monthly fee.
Sources
- Indeed search data for the Prompt Engineer title, and the Microsoft survey: Fortune, 7 May 2025
- Microsoft 2025 Work Trend Index, 31,000 workers in 31 countries: Microsoft WorkLab
