Priscila Andre de Oliveira is a senior software engineer at Sentry, an open-source maintainer, and a Vienna-based JavaScript community organizer who champions comprehension-first AI-assisted engineering. Her approach puts developers’ understanding of production software ahead of the speed at which agents generate changes.
Earlier in her career, Oliveira worked as a React developer at Eversports and participated in Vienna’s JavaScript meetups. After meeting contributor Juan Picado, she began improving Verdaccio’s interface, adding search, autocomplete, internationalization, and frontend tests. She became a core contributor and maintainer of Verdaccio, an open-source private npm registry, and helped organize ViennaJS.
At Sentry, Oliveira helped modernize a large, continuously evolving production application. In 2021, she and Mark Story explained the company’s incremental TypeScript migration, which converted approximately 1,100 frontend files and 95,000 lines of JavaScript without pausing product development. She later coauthored an account of Sentry’s migration from Enzyme to React Testing Library, encompassing roughly 5,000 tests and prioritizing observable user behavior over component implementation details.
How she approaches AI-assisted engineering
- Comprehension before code generation. An analysis of 116 of Oliveira’s AI sessions found that 67% involved understanding existing software, while 2% focused on generation. She uses agents to investigate regressions, reconstruct product decisions, navigate unfamiliar architecture, and gather context for code reviews.
- Catch Me Up. Oliveira built a personal repository-comprehension skill that structures investigation around architecture, conventions, features, execution traces, syntax, testing, and history. Applied to an unfamiliar Sentry integration-testing repository, it determined that tests intercept real Sentry envelopes instead of simulating them.
- Human judgment before implementation. Oliveira inserts an explicit comprehension checkpoint between an agent’s research and its implementation plan: engineers must evaluate what the agent found, correct mistaken assumptions, and understand proposed changes before shipping. Her AI Engineer Europe presentation ties that discipline to maintaining production reliability, reducing TypeScript technical debt, simplifying code, and removing obsolete feature flags.
