• Good foundation is still key
  • AI is a signal amplifier, not just a tool. It will amplify the good and the bad
  • System design is still very important (some say that hyper-generalists will survive if they have this skill)
• The bottleneck of writing code is now gone
  • This allows shipping faster, perhaps at the cost of learning and developing judgement for more junior people
  • Bottleneck is going up, for now it's at reviewing lots of generated code, it might go up further in the future
• Writing code might become a hobby
  • The same way AI can be better than most people at chess, but people still enjoy playing chess
  • Or how craft show makers might be hipster, but not the main trend
  • The notion of good code might change, good code used to be about nicely readable for humans, will this matter as much?
  • The people who loves the craft of writing code are quite disappointed and sad (identity crisis)
• The job of making software will become what the agents can’t do
  • Code is now implementation details, as agents can do that well
  • Elegance and judgement remain key human traits that are harder to delegate
• We're tending to be more of Product Engineers
  • The how is less important, but the what and the why still are
  • Good engineers were already good at product
  • Engineers that just take specs and do the code to do the spec and that's will disappear
  • What could this mean for Product managers?
• What made good managers might be what makes great engineers in the future
  • Context switching between tasks
  • Clear communication for writing specs for agents
  • Many parallels between managing (junior) people and many agents to divide work (project management)

What has changed in the role itself

• It won't replace us, but changes our role a lot: Why AI still won't replace software engineers
• It's still input -> output, but different to deal with as AI outcome is not deterministic
• Computer Science in the AI Era

Bottleneck change

Code is almost just implementation details now.

Code was never the bottleneck. It's making the correct decisions. This is changing now, as we're no longer constrained on weeks of implementation.
Will Larson says in Blog - Roadmap decisions rather than dates:

Software is not, as some argue today, free, but it’s no longer constrained on implementation weeks but instead on correct decisions.

Before, we were spending time thinking about what to do, and while it's getting built, we then think about what to do next. The speed of building things bought us time to think about what to do next, but now that got shorter so it's mainly just thinking about what to do, and documenting that?

Code review as bottleneck

Lots of code gets written.

Some say that's temporary, agents are becoming good at reviewing, it's all a matter of risk (low risk = auto review+deploy)

Moving more towards Product

Are we all Role - Product Engineer now?
Another path is moving closer to the client (FDE - Forward Deployed Engineer)

Output expectations are exploding

Feelings

2 types of engineer - where they get satisfaction from
The people who love the craft itself are more impacted.

Future

Coding manually might become a hobby

How judgement will develop

Software engineers need judgement, and the way the senior engineers built it has now disappeared.

Code review will die

---

Related:

• Identity as a software engineer
• Impact of AI in thinking
• Impact of AI on learning
• How to keep control of your codebase in the AI era
• Impact of AI in team collaboration
• Impact of AI in team formation
• Impact of AI on writing code
• Impact of AI on productivity
• Is software engineering still a safe job post AI