What's this about?
A debate in the Chinese AI community about Forward-Deployed Engineers. dotey's dark take: FDE is the model labs' open conspiracy — send engineers to help enterprises build agents (selling tokens), distill the enterprise's knowledge into Skills, then internalize those Skills into the model itself, after which the enterprise needs far fewer people. bozhou_ai's constructive counter: run FDE so every project works for the next one — structure every client interaction into a compounding knowledge base, making the delivery system, not any star employee, the asset.
What exactly is an FDE?
A Forward-Deployed Engineer is an engineer a software/AI company embeds directly inside a customer's organization to build working solutions on-site — a model popularized by Palantir and now adopted by AI labs (Anthropic, OpenAI) to make agents actually work in messy enterprises. The dual nature is the point of the debate: to the customer they're consultants; to the lab they're also a pipeline that converts private enterprise know-how into reusable Skills and, eventually, model capability.
What's the community saying?
- The extraction reading — dotey: if AI-driven efficiency doesn't grow the business, it just becomes headcount reduction; AI-literate engineers survive longest precisely by doing FDE.
- The flywheel reading — bozhou_ai: the compounding delivery system lets every ordinary team member wield the best member's judgment; organizational capability stops resetting when people leave.
- The shared pessimistic undertone — both takes accept the same premise: enterprise knowledge is being converted into model capability, and the transition will be rough on services labor.