LLMs and Agents: The Rise of AI Engineers

With the arrival of transformers and LLMs, the boundaries between data science and machine learning engineering are shifting. The role of AI Engineer has emerged between GPT-4 and GPT-5, while the Forward Deployed Engineer is spreading in the AI context. Agentic tools are automating entire actions and bringing these profiles closer to product challenges.

Agentic Automation: End-to-End Actions Executed

Tools like Codex, Cursor, and recent ChatGPT plugins automatically perform tasks such as checking emails, opening merge requests, bug verification, and writing entire codebases. They are termed agentic because they carry out actions. These uses differ from employing ChatGPT in 2023 and 2024 to draft an email or fix a bug, which is described as a productivity gain rather than complete automation. The acceleration is thus shifting from occasional assistance to automated action chains.

Missions Shifting Towards Product

Coding is no longer presented as a major obstacle for Data Scientists. In this context, Data Scientists and Machine Learning Engineers are described as moving closer to the responsibilities of Product Managers, with more time devoted to identifying problems, designing systems, and evaluating them.

What the Role of AI Engineer Entails

Between GPT-4 and GPT-5, the position of AI Engineer has emerged. This profile masters language models and their operational use: designing and fine-tuning prompts, selecting suitable models, implementing safeguards, anticipating hallucinations, evaluating outputs, and ensuring reliability for production. A background in data science or experience in training models is not required, although these skills are beneficial.

The Forward Deployed Engineer in the AI Context

The term Forward Deployed Engineer was first introduced by Palantir, where engineers worked directly with clients to identify complex problems and design solutions. Its use in AI now differs from this origin: organizations unfamiliar with the technology seek interlocutors capable of precisely identifying needs and proposing executable designs. In AI, the role is described as highly hybrid, combining the technical depth of an AI Engineer or ML Engineer, the analytical thinking of a Data Scientist, and the product intuition of a Product Manager.

Historical Milestones: From Transformers to LLM Products

In 2017, a Google paper introduced the transformer architecture. This serves as the foundation for language models like GPT. Practitioners' attention grew with GPT-3, and GPT-4 is described as a turning point that convinced companies to build real products. The timeline mentions GPT, GPT-2, GPT-3, GPT-4, GPT-5, and GPT-6 Astra, the latter being slated for release on October 3, 2026.

Changes Compared to Historical Roles

The initial landscape contrasted a Data Scientist close to the product, relying on statistics and tools like Python and SQL, with a Machine Learning Engineer focused on training pipelines, neural networks, performance, and deployment, using Python and libraries like TensorFlow or PyTorch. The Applied Scientist, popularized by companies such as Amazon and Microsoft, was closer to engineering by making code production-ready. Current products like ChatGPT, Codex, or Claude rely on the orchestration of multiple LLMs and are part of an agentic dynamic oriented towards automation. The proposed summary asserts that the emergence of AI Engineers stems from transformers and LLMs, and that agentic AI has reduced coding as a bottleneck by bringing technical roles closer to product challenges.