Comprehend First, Code Later: The AI Skill I Rely On Daily — Priscila Andre de Oliveira, Sentry

TL;DR

• 1

  Most AI value came from comprehension, not generation — After analyzing 116 Claude sessions, Priscila found 67% of her usage was comprehension and only 2% was code generation.

• 2

  She built a personal “catch me up” skill to explore unfamiliar code quickly — The local Markdown-based prompt organizes questions into six modes: architecture, convention, feature trace, syntax, testing, and history.

• 3

  Sentry’s scale makes understanding non-optional — The company has 15+ years of code, 100k organizations depending on it, and roughly 100 PRs merged every day, so mental models go stale fast.

• 4

  AI helped her remove waiting and manual archaeology from daily work — Instead of git blame, digging through Slack, or waiting on teammates in other time zones, she now gets regressions and product-decision context in seconds.

• 5

  Sentry is going all-in on internal AI agents, but still pairing that with code quality work — She names projects like Abacus, Warden, and Junior, while also pointing to a three-month “quality quarter” spent paying down technical debt.

• 6

  Her workflow adds an explicit understanding step before planning and implementation — Referencing Jack Nation and Rich Hickey’s “simple made easy” framing, she argues you must understand what the agent found before letting it plan or write code.