SWE-bench co-creator designs Stanford CS329Z HW2 on Terminal-Bench-style agent evals
FinanceYF5 · x · 2026-11-16
CS329Z Assignment 2 focuses on agent evaluation, designed by SWE-bench co-creator and course instructor John Yang. Built on the Harbor framework, it teaches students to build Terminal-Bench-style evals that reflect real tasks — judging not just whether agents finish tasks but how well. The course is also supported by Laude Institute, Thinking Machines, Google Cloud, AWS and Harbor Framework, which provide students API credits to run and evaluate agents hands-on.
More from coding & agent
- json-render for Swift brings native Generative UI to iOS and macOS — amos_gyamfi · 2026-11-16
- Sentry CEO tries tenant SQLite on Durable Objects, back to Postgres in 30 minutes — zeeg · 2026-11-16
- Unified 27B local agent model open-sourced: vision, typed decider, abliterated, runs on 16GB VRAM — ButtercupLyn100 · 2026-11-16
- Stanford CS329Z draws 100+ students in debut, publishes full syllabus with industry API credits — FinanceYF5 · 2026-11-16
- Stanford CS329Z open-sources HW1: build an agent harness from scratch for under $5 — FinanceYF5 · 2026-11-16
- Stanford opens CS329Z agent engineering course: build a Claude Code-like agent for under $5 — FinanceYF5 · 2026-11-16