EO StudioFrom Writing Code to Managing Agents. Most Engineers Aren't Ready | Stanford University, Mihail Eric
At a glance
WHAT IT’S REALLY ABOUT
AI-native engineers manage agents, not code, and need fundamentals
- Junior software engineers face a tougher job market due to post-2021 overhiring, layoffs, and a surge in CS graduates converging with AI productivity gains.
- An AI-native engineer combines traditional programming fundamentals with competence in agentic workflows, where a single developer increasingly manages multiple AI agents.
- Top practitioners scale agent usage incrementally, carving work into isolated tasks and mastering rapid context switching across parallel agent threads.
- Agent-friendly codebases rely on explicit contracts—tests, consistent documentation, linting, and consistent design patterns—to prevent agents from compounding errors into spaghetti code.
- Beyond functional output, “incredible” software comes from developed taste and continuous experimentation, while AI-native organizations win by allocating intelligence and embedding AI into products.
IDEAS WORTH REMEMBERING
5 ideasThe junior job crunch is a multi-factor ‘perfect storm.’
Layoffs after 2021 overhiring, a larger supply of CS graduates, and employer substitution toward fewer AI-native hires together make entry-level roles harder to secure.
AI-native engineering is management plus fundamentals, not prompts alone.
The differentiator is strong system design and programming foundations paired with the ability to run and supervise agentic workflows effectively.
Don’t start with 10 agents—earn scale one workflow at a time.
Add agents only when you can clearly isolate tasks and confidently specify boundaries; otherwise coordination overhead and error interactions can degrade the system.
Context switching becomes the “last boss” skill in multi-agent work.
You must quickly reconstruct what each agent was doing, what assumptions it made, and what’s blocked—mirroring the core competence of effective human managers.
Tests are the contracts agents can reliably follow.
Without robust test coverage, agents lack crisp definitions of correctness; outdated READMEs and ambiguous behavior create conflicting guidance that derails changes.
WORDS WORTH SAVING
5 quotesA single developer become a manager of agents.
— Mihail Eric
Adding more agents doesn't always create for a better system. In fact, it can make for a lot worse systems, actually, if, if you just let them go and do whatever they want.
— Mihail Eric
Agents only can operate on contracts, like explicitly defined contracts of software.
— Mihail Eric
Experimentation is sort of the name of the game in becoming an AI-native software developer.
— Mihail Eric
I think we're in a world where increasingly what matters is your ability to allocate intelligence.
— Rem Koning
High quality AI-generated summary created from speaker-labeled transcript.
