Lemire's Sedgwick Lecture to unpack what actually makes agentic coding work: tests, context, parallel agents

lemire · x · 2026-09-23

Daniel Lemire shared details of his Sedgwick Lecture, "Programming in the Era of AI." Key claims: tools like Cursor, Claude Code, and GitHub Copilot have led many leading developers to stop typing code manually; agentic AI now designs, writes, tests, and deploys apps from high-level requirements; frontier models hit superhuman performance on parts of SWE-bench Verified, and open-source models make this cheap. The talk will focus on practical components that make agentic approaches effective—benchmarks and tests as guiding constraints, context management, and orchestrating parallel agents—and on what to re-teach, re-organize, and re-research.

Original post →

More from coding & agent

coding & agent channel →