Stop hiring AI Engineers. Hire platform engineers who can read papers.

Open any AI Engineer job description posted in the last 12 months. The requirements look like this. PhD or MSc in machine learning, computer science, or related field. Experience with PyTorch, TensorFlow, JAX. Familiar with transformer architectures. Published research a plus.

Then look at what the role actually does. Wire up an API call to OpenAI. Set up a vector database. Build a RAG pipeline. Deploy a model behind a load balancer. Monitor it in production. Roll it back when something goes wrong.

Most companies do not need AI Engineers. They need platform engineers who can read papers. The distinction matters. A platform engineer who can read papers can pick up the relevant technical context for any AI project in a few weeks. A researcher who has never deployed to production cannot pick up infrastructure work in the same timeframe. Production engineering takes years to learn. Reading papers takes months.

Here is the job description I would write for the role most companies actually need.

• 5+ years building and operating distributed systems in production
• Deep experience with at least one cloud provider, including IAM, networking, and cost controls
• Comfortable reading research papers and translating them into production patterns
• Hands-on with at least one of: vector databases, embedding pipelines, LLM API integration, model serving infrastructure
• Has been on-call for systems they built. Has been woken up. Has written the runbook.
• Curious about ML, but not defined by it

Notice what is not on this list. No PhD. No research publications. No specific framework experience. No requirement to have trained a model from scratch.

The reason is simple. The hard parts of AI in production are not the model. They are everything around it. Cost attribution. Latency budgets. Cache invalidation. Failure modes. Rollback procedures. Observability. Compliance. The model is a black box you call. The platform around it is the thing you actually have to build.

If you hire a researcher to do this work, one of two things happens. They struggle for six months and leave for a research role. Or they stay and build something that works in the notebook and fails in production. Either way, the company loses.

If you hire a platform engineer who can read papers, they have shipped to production before. They know what breaks. They know what to monitor. They know what the on-call experience needs to feel like. They can read the latest paper on retrieval augmented generation and tell you, within an hour, whether it is relevant to your problem and what it would cost to ship.