Reading code: the superpower of this decade

Lessons learned 5/11 on Cursor after 15 years in software

AI pushes developers into a new paradigm where generating code is faster than thinking about what you'd want to do.

You can quickly find yourself in a situation where code writes itself faster than you can understand it. It's then fundamental to know what method the AI employs, because biases can appear with every choice.

The new superpower

The new superpower of this decade is undoubtedly knowing how to read code and understand its structure. That's what allows you to iterate quickly while maintaining satisfactory quality on your project.

The importance of logs

The other important reading is your logs during execution of your software. Unexpected behaviors usually leave a trace, even when execution runs smoothly.

For example, AI has an annoying tendency to add fallbacks in certain functions, where you'd prefer a clear, massive error so you can fix the problem at its root.

A try/catch that silently swallows an error is a ticking time bomb. AI loves putting them everywhere because it "avoids crashes." Except you'd rather have an explicit crash than a silent bug that corrupts your data for three weeks before anyone notices.

Summary

• Learn to read code — PRs, frameworks, colleague's code, AI-generated code — to quickly identify the structure of generated code
• Systematically review error logs and alerts to detect unexpected behaviors
• Beware of silent fallbacks: explicitly ask the AI to throw errors rather than masking them

AI writes fast. Your job is to verify that it writes right.

---

Originally published on LinkedIn.