Ben Burtenshaw

Activities

historical

Ben Burtenshaw is an AI engineer at Hugging Face with a background in AI systems engineering. He has worked on making Hugging Face’s hub infrastructure more useful for practical AI engineering workflows, including kernel publishing, skills systems, and automated evaluation.

present

Burtenshaw is a leading voice on the thesis that coding agents should tackle hard AI systems engineering problems — not just application-level tasks. He presented this work at the AI Engineer conference (May 2026), demonstrating three escalating use cases: agent-assisted CUDA kernel writing, zero-shot fine-tuning on the Hugging Face hub, and the AutoLab distributed multi-agent research system.

He introduced the Upskill tool for generating skills (file-based context with examples and tests), generating evaluations, and comparing models on the same task to optimize cost and quality. He recommends Trackio as the monitoring layer for multi-agent systems, favoring it for its openness and raw data access over opaque dashboards.

His practical guidance: agents work best with open primitives and verifiable tasks. Skills convert zero-shot tasks to few-shot by giving agents examples, scripts, and tests — dramatically improving reliability.

Connections to other people and companies

• AI Engineer at Hugging Face
• Developer of AutoLab and Upskill tools
• Speaker at AI Engineer conference 2026

Interests

• AI systems engineering and CUDA kernel optimization
• Multi-agent research automation
• Skills as a portable agent context abstraction
• Hugging Face hub as infrastructure for AI engineering workflows
• Open, benchmark-driven development

Sources

Source: AI Engineer YouTube — “Your Coding Agent Should Do AI System Engineering — Ben Burtenshaw, Hugging Face” (May 2026).