Comprehend First, Code Later — The AI Skill I Rely On Daily
- Channel: AI Engineer (aiDotEngineer)
- Speaker: Priscila Andre de Oliveira, Senior Software Engineer @ Sentry
- Event: AI Engineer Europe 2026
- URL: https://www.youtube.com/watch?v=li0SaBt9RDM
- Tags:
#AI#CodeComprehension#Sentry#AIEngineer#Productivity
Overview
Priscila Andre de Oliveira (Sentry) analyzed 116 of her own Claude sessions from daily work and made a counterintuitive discovery: 67% of her AI usage was for code comprehension, and only 2% was for code generation.
The most powerful AI skill isn’t prompting for new code — it’s using AI to understand existing codebases faster.
The Data
| Type | Share |
|---|---|
| Code comprehension | 67% |
| Other (debugging, planning, etc.) | 31% |
| Code generation | 2% |
Working in a complex, multi-service codebase at Sentry, she found herself repeatedly asking Claude to explain, trace, map, and summarize existing code — far more than asking it to write new code.
Key Takeaways
- Comprehend first, code later — understanding the system is the bottleneck, not writing code
- AI’s biggest impact on senior engineers isn’t writing — it’s reading at superhuman speed
- The prompts were repetitive (same comprehension questions to different parts of the codebase)
- Code gen is a tiny fraction of practical daily AI use for experienced engineers
Why It Matters
This flips the common narrative. Most AI product positioning focuses on code generation (“write 10x faster”). But the real leverage for experienced engineers is comprehension — understanding a 10-year-old monolith, tracing a bug through 15 services, or onboarding into an unfamiliar codebase. The faster you understand, the faster you act.
For junior engineers, this suggests the skill to build isn’t “prompting for code” — it’s learning to ask the right comprehension questions.