How do I get a job in AI without a PhD?

Yes — and the route is wider now than it was, because most AI work in industry is engineering rather than research. PhDs are needed to *create* new model architectures. Almost nobody is doing that; the vast majority of paid AI work is applying existing models to real problems, which is a software and data problem. The roles that don't require a PhD, roughly in order of accessibility from a software background: 1. **AI application engineer / AI engineer.** Building products on top of existing models — retrieval over company documents, tool use, structured outputs, evaluation harnesses, cost and latency management. This is mostly excellent backend engineering plus knowing how these systems fail. Highest demand, lowest barrier from where you are. 2. **Data engineer.** Pipelines, storage, quality, feature availability. Every ML team is bottlenecked here and it's chronically undersupplied. No research required at all. 3. **ML engineer.** Training, deploying and maintaining models in production — monitoring, retraining, serving, drift. More engineering than mathematics in practice. 4. **Data scientist / analyst.** Statistics, experimentation, classical ML on business problems. A PhD helps in some companies and is genuinely unnecessary in most. 5. **MLOps / platform.** Infrastructure for the above. What to actually build to get hired: - **One substantial end-to-end project, deployed.** Not a notebook — a working system with an interface, real data, and a live URL. A retrieval system over a document corpus, or a fine-tuned model serving predictions through an API. This is worth more than five Kaggle notebooks. - **Evidence you can evaluate.** Show that you measured whether your system was actually good, and what you changed when it wasn't. Say what failed. This is the scarcest skill and the strongest signal — most applicants show only the version that worked. - **Solid software engineering.** Version control, testing, deployment, clean code. A lot of self-taught ML people are weak here, and it's frequently the actual reason for rejection. - **The fundamentals, properly.** Overfitting, validation strategy, metrics beyond accuracy, data leakage. You will be asked, and shallow answers are obvious. The positioning that works best: come in through an adjacent role at a company doing AI work, then move internally. A backend engineer who volunteers for the ML integration work is on a much shorter path than an outsider applying cold to research-flavoured listings. And read job descriptions carefully — 'PhD preferred' on a listing that describes building pipelines and APIs is aspirational boilerplate, and people without one get hired into those roles constantly.