From Writing Code to Managing Agents. Most Engineers Aren't Ready | Stanford University, Mihail Eric - TooLong.XYZ

Detailed Summary

A new class of 'AI-native' engineers is emerging where the primary role shifts from writing code to managing agents.
Mihail Eric introduces his Stanford class, 'The Modern Software Developer,' which is the first to focus on AI across the entire Software Development Life Cycle (SDLC).
The high demand for this knowledge was evidenced by the class filling up with over 100 students within hours of announcement.

Junior engineers are struggling; anecdotes suggest graduates applying to 1,000 positions may only hear back from two.
Three factors created a 'perfect storm': the 2021 overhiring surge followed by massive layoffs, a 2x to 3x increase in CS graduates over the last decade, and the rise of AI allowing companies to hire fewer, more efficient people.
The current generation must be the first to be fully AI-native while maintaining strong fundamentals in system design and algorithms.

AI-native engineers use 'agentic workflows' but should build them up piecemeal rather than trying to manage 10 agents at once immediately.
Success involves isolating tasks (e.g., one agent for a logo fix, another for header copy) and adding complexity only when confidence is high.
A core skill is 'context switching'—the ability to monitor multiple agents in a terminal or IDE and provide meaningful direction when they get stuck, similar to a human manager.

Agents require 'contracts' to operate correctly; if a codebase lacks test coverage, the agent has no way to verify its work.
Consistency is vital: if the README contradicts the code, or if two different APIs are used for the same task, the agent will become confused and produce errors.
Agents compound errors quickly; a single misunderstanding in step one can be magnified in subsequent iterations, leading to 'spaghetti code.'
Robust linting, style checking, and consistent design patterns are essential to keep agents on the right track.

The difference between functional and incredible software is 'taste,' which is developed by doing extra work beyond the basic requirements.
Top engineers are those who continue building and refining their projects into startups even after a class or assignment ends.
Continuous experimentation is required; for example, the Anthropic team reportedly rewrites parts of Claude every week using Claude itself.

Senior developers can be stubborn and resistant to AI tools due to 20 years of ingrained habits.
Junior engineers act as 'sponges,' lacking the 'scars' of industry difficulty, which allows them to tackle problems with a brave, startup-oriented mindset.
Computer Science is increasingly about 'learning how to think' and breaking down complex systems, similar to mathematical problem-solving.

Rem Coning highlights that the future belongs to those who can 'allocate intelligence' effectively.
Being AI-native means embedding AI into the product so it interacts directly with the customer, removing the human from the loop.
The next trillion-dollar companies will likely emerge from solving how AI agents collaborate and communicate with one another.