News

Stanford's CS146S brings open-source maintainers in as reviewers, Latent Space reports

Published 2 min readAI-assisted · human approved
TL;DR

In the second run of Stanford's CS146S, instructor Mihail Eric had open-source maintainers review student code, according to a Latent Space podcast posted Nov 8. Eric also leads AI at Monaco.

Why it matters

If coding agents can finish standard homework, courses and hiring loops need other ways to test skill. Reviewing and defending a real patch is one approach teams can borrow for onboarding or training.

Original source

Watch on YouTube · Loads from youtube-nocookie.com only after you press play

What happened

Latent Space published a 121-minute podcast episode on Nov 8 with Mihail Eric. He teaches The Modern Software Developer at Stanford and leads AI at Monaco. The episode was recorded to mark the AI Engineer YouTube channel passing a million subscribers, according to the show's description.

The main topic is the second run of Stanford's CS146S. Eric added open-source maintainers to the course. According to Latent Space, students must get up to speed on a codebase they have not seen before, draft a proposed change, and then talk the patch through with the person who could merge it.

The framing question is what to give students when a coding agent can finish last year's homework from a single prompt.

Key details

  • The episode was posted Nov 8 and runs 121 minutes.
  • Eric opens the CS146S syllabus on the show. It is ten weeks long.
  • He says which lessons held up on his own engineering team, per the show notes.
  • Monaco built its agent workspaces, bought its code review tools, and gave an on-call agent access to Datadog and GitHub.
  • The hosts ask how Eric decides which tools his team should own. One timestamped section is titled "You buy to learn, and you build to keep."
  • Other segments cover the loop, tools and harness inside a coding agent, context engineering for a large codebase, and making repos agent-ready.
  • Later sections cover code review as a bottleneck, security and permissions, background agents, and Stripe's Minions.
  • The show notes link a from-scratch coding agent build and a walkthrough of a real agent harness, both by Eric.

Why it matters

The episode describes two things that apply beyond a classroom. One is a test of engineering skill that a single prompt does not settle: reading unfamiliar code and defending a change to a reviewer. The other is how one team splits its tooling between building and buying.

This is a podcast description and show notes, not a published curriculum or study. The details above come from the listing on Latent Space. The full discussion is in the audio and video.

FAQ

What is Stanford CS146S? It is a Stanford course with a ten-week syllabus and assignments, taught by Mihail Eric. The Latent Space episode covers its second run.

What changed in the second run of the course? According to Latent Space, Eric brought open-source maintainers into the course to review student work.

What does Monaco's on-call agent have access to? Per the show description, Monaco gave an on-call agent access to Datadog and GitHub.

Checked claims

ClaimStatusEvidence
The second run of Stanford's CS146S brought open-source maintainers into the course.confirmedIn the second run of Stanford’s CS146S, Mihail Eric brought open-source maintainers into the course.
Mihail Eric teaches at Stanford and leads AI at Monaco.confirmedwho teaches The Modern Software Developer at Stanford and leads AI at Monaco
Monaco built its agent workspaces, bought code review tools, and gave an on-call agent access to Datadog and GitHub.confirmedMonaco built its agent workspaces, bought its code review tools, and gave an on-call agent access to Datadog and GitHub.
The CS146S syllabus runs ten weeks and includes assignments.confirmedStanford CS146S — The ten-week syllabus and assignments.
One segment of the episode covers open-source maintainers reviewing the students' code.confirmed10:02 Open-source maintainers review the students’ code

· Editorial policy

Question for readersIf a coding agent can finish your standard exercises in one prompt, what would you ask a junior developer to do instead?

Discussion · 0

No comments yet. Start the discussion.

Sign in to join the discussion.