Lyft lets domain experts build customer-facing agents: guardrails, prompt linting, trajectory evals
bigdata · x · 2026-10-22
Akshay Sharma (ML Engineer) and Nick Ung (Staff Applied Scientist) at Lyft explain on The Data Exchange podcast how their platform lets domain experts—not just engineers—create customer-facing AI agents, while engineers supply the underlying tools, guardrails, and evaluation infrastructure. Key practices include prompt linting, guardrails, evaluating agents with simulations and synthetic users, scoring entire agent trajectories rather than just model outputs, and turning production traces into critical training data. The interview also covers multi-agent systems, memory, safety, open-weight model routing, and shared eval infrastructure.
More from coding & agent
- swyx hails Opus 5.5 as best work-estimator model; dev counters that review time is the real bottleneck — petrusenko_max · 2026-10-23
- Agent Field Kit launches free agentic defense toolkit against LLM-enhanced attackers — alvelda · 2026-10-23
- LiteLLM PyPI compromise and Railway breach: Oct 21 attack timeline and free response kit — alvelda · 2026-10-23
- Daily Experiment Planner agent: one prompt to save scientists hours of lab prep — nlarusstone · 2026-10-23
- Two £300 Snapdragon mini PCs, no GPU: 46 tok/s local Qwen3.8 Flash Next with custom ToRc engine — WorldlyAd7946 · 2026-10-22
- Chrome engineer asks: do agents improve incident response or just ship unpatched deps 10x faster? — fhinkel · 2026-10-22