Three Java Champions interviewed by GitHub say AI coding tools are most useful when developers provide project context and remain responsible for every result. Their examples cover production software, open-source maintenance, modernization, and documentation.
Vlad Mihalcea said Copilot helped him build a previously postponed application in two days instead of the weeks he expected. He still reviews generated plans and code, and uses custom instructions derived from his earlier writing and coding practices so output better matches established conventions.
The developers also described choosing different models for different jobs rather than sending every task to the most expensive option. Deeper design work may justify a larger reasoning model, while reports, release notes, or documentation checks can use lighter models.
The examples are individual experiences, not a controlled productivity study. Their shared lesson is more practical: generated code must still be tested, reviewed for maintainability, and fitted to the existing system. Fast output can create more work later if it ignores architecture, security, or the conventions that future maintainers expect.