Thomson Reuters' CoCounsel FDE ads make LLM evals core work, and the Senior role asks for legal training

FDE Times compared the job ads for Thomson Reuters' CoCounsel Forward Deployed Engineering team. The mid-level role builds and runs RAG pipelines, agentic workflows and multi-component AI systems, and also helps evaluate LLM output on legal accuracy, relevance, citations, consistency and hallucination. At Senior level the wording moves from helping with evaluation to leading it.

The standout requirement is in a Senior ad posted on Dice: candidates need a JD, formal legal training, or equivalent deep experience in a legal practice setting. The article ties this to Thomson Reuters' own CoCoBench benchmark, which grades output by whether a practising lawyer would find it acceptable, so whoever writes the evals has to know how lawyers read an answer.

Why it matters for FDEs: in vertical AI, proving output meets a profession's standard is becoming as important as building the pipeline. The piece suggests a portfolio project of a RAG system on public legal documents with its own eval suite that scores retrieval, answer accuracy, citation accuracy and fabricated sources separately.