Three Java Champions with more than a decade of experience have shared how they use GitHub Copilot in production and open-source work. Their examples range from building features and migrating teaching materials to investigating bugs. Across their approaches, they emphasise that AI can accelerate execution, but developers must still provide context, verify the output and own what gets released.
Vlad Mihalcea uses Copilot in IntelliJ IDEA to investigate code and prototype features for Hypersistence Utils and Hypersistence Optimizer. He said a web application for viewing performance-optimizer event history, a project he had previously expected to take weeks, took two days to implement with Copilot. He still reviews both the proposed plans and generated code, and uses custom skills to align documentation with his writing style and existing project conventions.
Patrick Baumgartner described three different uses. In one, Copilot helped migrate about 900 slides across 15 presentations from Reveal.js to Marp. The work took roughly eight to ten hours. Baumgartner checked the results using screenshots, visual and algorithmic comparisons, side-by-side animation reviews and PDF comparisons, while retaining responsibility for the content and teaching intent.
In another case, a user reported that the Enter key did not work in some forms of a terminal application. Baumgartner used Copilot to help reproduce the issue with UI tests, identify the underlying problem in the Tamboui project and submit a fix that was merged. He also used the tool to find a more compact way to represent test steps when the original descriptions exceeded the available context.
His third example was a mathematics-learning app designed around the Swiss school curriculum. Working with Copilot, he explored learning approaches and reconsidered his initial plan to make the app universal. The resulting application supports German, French, Italian and English, and includes practice and two-player modes. It has no backend, so user data remains on the device.
Baumgartner said he now mainly uses agentic workflows, choosing an editor when he needs close visual feedback and a terminal-based coding agent for larger tasks. He also recommended discussing requirements before asking an agent to implement a substantial change, reviewing a plan, and setting task-specific checks to prove the work is complete. He and Mihalcea both described reusable skills as a way to carry project conventions and lessons from earlier work into future tasks.
Their accounts frame Copilot as a tool for accelerating work and challenging ideas, rather than a replacement for engineering judgement. Baumgartner’s guiding principle was to delegate execution, not responsibility; he said developers still need to inspect and test the results themselves.
Sources: GitHub Blog.