
[One Million AI Engineer Special] How Stanford Teaches AI-Native Engineering — Mihail Eric, Monaco
Latent Space: The AI Engineer Podcast
- 2h 0m
- Transcript
Summary
Written by a model from the transcript; not yet reviewed.
This episode covers the structure and evolution of the Stanford course CS146S, 'The Modern Software Developer,' taught by Mihail Eric, who is also the head of AI at Monaco. The discussion traces how the curriculum shifted from broad exposure to AI development tools last year to a rigorous, practitioner-focused deep dive this year. They explore the mechanics of coding agent harnesses, context engineering at scale, open-source code contributions, and the realities of running AI-native software in production. The conversation also details Monaco's approach to GTM sales platforms, agentic workflows, and the broader organizational challenges of code reviews, security, and the emerging software factory.
- Developers entering the modern workforce must master foundational software design and architecture because AI agents have raised the performance floor for junior engineers. 1:08
- To truly understand how modern coding agents operate under the hood, use MITM proxy to inspect system prompts and message structures. 27:53
- Developers working with massive codebases must move beyond vanilla prompting and adopt structured context research and proposal frameworks. 33:17
- Engineers should systematically prune and refine their custom agent skills to prevent unwanted triggers during different project tasks. 44:05
Chapters
- 0:00Episode Celebration and Guest Introduction
Hosts celebrate 1 million subscribers and introduce Mihail Eric, Stanford professor and AI lead at Monaco.
- 0:58History and Motivation of CS146 Course
Mihail explains the genesis of Stanford’s CS146, its evolution, and why it was created to formalize AI‑native engineering education.
- 3:45Monaco AI Practice and Curriculum Updates
Mihail contrasts last year’s broad survey of coding agents with this year’s focus on context engineering, architecture, and practical tooling.
- 4:33Future Directions and Experimentation Philosophy
Mihail emphasizes continuous experimentation, staying current with evolving models, and the need for hands‑on practice to master AI‑native engineering.
- 7:03Tooling, System Prompts, and Model‑Harness Trade‑offs
Exploration of how system prompts evolve, the balance between model capabilities and harness design, and the implications for developers.
- 8:00Context Engineering and Multi‑Agent Management
Deep dive into context management, sub‑agent inheritance, and strategies for scaling prompts across large codebases.
- 10:01Student Experience and Open‑Source Integration
Discussion of student demographics, open‑source partner involvement, and the shift from simple assignments to real‑world OSS contributions.
- 12:44Open‑Source Contributions and Real‑World Projects
Mihail highlights the importance of OSS contributions, mentorship, and real‑world projects in building practical AI‑native engineering skills.
- 15:15Curriculum Breakdown: Ten Weeks of AI‑Native Engineering
Mihail outlines the ten‑week syllabus, covering harness construction, context engineering, agent design, security, and factory‑scale workflows.
- 39:19Interactive Skill Learning and Quiz Design
Explores how interactive quizzes reinforce understanding of AI skills and the importance of active engagement.
- 40:36Web Interaction Skills and Best Practices
Conversation about skills that interact with the web, best practices, and the importance of understanding the code base.
- 42:56Skill Creation and Maintenance Challenges
Discussion on writing effective skills, the pitfalls of poorly written triggers, and the necessity of pruning unused skills.
- 44:05Monaco AI Platform Overview
Introduction to Monaco, its all‑in‑one AI‑native sales platform, and its focus on early‑stage startups.
- 51:40Monaco’s In‑House Tooling and Agent Ecosystem
Details Monaco’s use of in‑house tools like Monocoder and Hal, and how they integrate with external vendors.
- 1:08:48AI‑SRE and Enterprise Scale Challenges
Hosts discuss the role of AI‑SRE in managing large Kubernetes infrastructures and the complexity of enterprise systems.
Highlights
- 1:08
First generation of AI engineersinsight
The guest explains that the current entering cohort of junior engineers is the first generation experiencing the profound shift brought by AI-native software development.
- 36:34
The dumb zone in contextinsight
The hosts discuss the importance of managing context windows carefully to avoid letting agents wander into the 'dumb zone' where they make avoidable errors.
- 39:03
Forcing active engagement with skillsstrategy
The guest highlights an educational skill called Explain Diff that finishes with a quiz to ensure developers actually understand the code changes rather than tuning out.
- 58:01
Stealth building an all-in-onestory
The guest explains that Monaco stayed in stealth for over a year to build an all-in-one GTM platform rather than starting as a narrow point solution.
- 1:01:58
Buy to learn, build to keepstrategy
The guest shares a core engineering principle at Monaco: they buy off-the-shelf tools to learn what works and build in-house when mature solutions do not exist.
- 1:10:35
Sandbox assignments for SREsstory
The guest describes a class assignment created by Resolve where students use an SRE agent to debug simulated multi-service infrastructure failures.
- 1:37:37
The God box infrastructure convergenceinsight
The host argues that in infrastructure engineering, all systems eventually converge into a single 'God box' that handles routing, security, and permissions.
- 1:54:31
The necessity of engineering psychosisdebate
The host asserts that engineers need to go through the intense cycle of over-engineering and trying every tool—'psychosis'—and come out the other side to build real products.
Quotes
“There is a way to do this that actually does scale out to codebases that are, you know, millions of lines long.”
Unnamed speaker · 34:35 “I think that when I reflect on what last year's version of the class was, I mean, coding agents had been around, but we were still in kind of that phase where companies were emerging that were tackling every single part of like the ISDLC.”
Unnamed speaker · 3:54 “I think there’s a lot of anxiety from students, as you can probably imagine.”
Unnamed speaker · 5:50 “This one, I think recently, which is like kind of, there's no situation in which turning your brain off is a good idea.”
Unnamed speaker · 39:39 “But no, you should write your skills yourself. And actually, when you try to write them naively, they're usually bad.”
Unnamed speaker · 42:45 “Why? Because it's badly written, because the trigger says like, oh, you call me whenever blah, blah, blah happens.”
Unnamed speaker · 44:18 “And so we are, I would argue are the, you know, the, the, I would, I would claim that among the best, if not the best, sort of like a startup focused, uh, sales platform.”
Unnamed speaker · 52:07 “But then you can backfill the, the model side once you have the user base.”
Unnamed speaker · 54:39 “The whole point of open source is that, you know, what is it, like sunlight is the best disinfectant?”
Unnamed speaker · 1:22:54 “If you don't know what the difference between off and an off Z is, please learn it.”
Unnamed speaker · 1:29:09
About this episode
What do you assign when a coding agent can finish last year’s homework in one prompt? In the second run of Stanford’s CS146S, Mihail Eric brought open-source maintainers into the course. Students have to learn an unfamiliar repo, propose a change, and ...
Transcript
26,916 words · 3 speakers
Topics
- Monaco
- MCP
- Claude Code
- Stanford
- Cloud Code
- AIE
- Opus
- MCP portals
- CS146
- Latent Space
Page updated · transcript and summary are machine-generated