From Writing Code to Managing Agents. Most Engineers Aren't Ready Stanford University, Mihail Eric (Category Engineering)

Interesting. interview Mihail Eric, Head of AI at Monaco and Stanford lecturer, where he teaches a course.CS146S: The Modern Software Developer" Mihail Eric talks about where software development really goes. The main idea is that the profession does not disappear, but the center of gravity goes from manually writing each line to orchestrating AI agents, decomposing tasks, verifying results and designing an environment in which agents can be trusted. CS146S: The Modern Software Developer.

In this interview 15The minute interview had some interesting thoughts.

1Eric explains why the junior market is so stormy

• Overheated hiring after. 2021 year and subsequent reductions A sharp increase in the number of CS graduates AI, which is making companies increasingly think not “who else to hire,” but “how many AI-native engineers it takes to close the same amount of work.”

2Eric explains who is an AI-native engineer This is not just a pro-industrial engineer, but rather a developer with a normal base in system design, algorithms and traditional development. This engineer should be able to work with agency workflow According to Eric, the main mistake is trying to run many agents at once. But it is better to start the other way around: first one agent per workflow, then gradually add only isolated tasks and only then scale the orchestration.

3Eric reveals agent-friendly codebase For an agent, tests are not a "nice to have" feature, but actually contracts. If there are few tests, README is outdated, and the same entities are created by different patterns in different places, the agent begins to guess - and very quickly begins to breed errors. As a result, texts, documentation and uniform database patterns are not just engineering hygiene, but simply the basis for AI-native development.

4Eric notes that multi-agent workflows are more like management than classical programming. You need to be able to quickly switch context, keep in mind what each agent is doing, understand where it's stuck, and gently get it back on track. In fact, a strong AI-native engineer is a bit of an agent team manager.)

5Eric highlights importance of food and engineering taste A functional product and a really strong product share not only the correctness of the code, but also the desire to go the “last mile”: make the feature more stable, useful, deeper, bring UX and robustness. Eric attributes this to constant experimentation: even the teams that build AI tools are constantly rewriting and reinventing their workflow.

In general, the author 15 He sold me his course on modern software development. whose lectures are already available on YouTube (However, this is not the official version, which I did not find, but generated via notebook lm based on course materials.)

#Agents #Software #Engineering #Management #DistributedSystems #SystemDesign #AI