Three Java experts explain how they use AI coding help at work

Three Java Champions, a title for respected Java developers, have described how they use GitHub Copilot, an AI coding assistant, in real work. Their shared message is that the tool saves a lot of time, but people stay responsible for what ships. All three have more than a decade of experience with Java.

Weeks of work done in days

One Champion maintains two Java tools, Hypersistence Utils and Hypersistence Optimizer. He had put off building a web app that shows how performance improvements play out over time, because it would have taken weeks. With Copilot, which he compares to a new team member, it took two days. The app will ship in the next version of Optimizer.

How they keep control

The same developer still reviews every plan and every piece of code the AI creates. In his words, the goal is to improve his productivity, not replace him. Sometimes the answer is not good enough and he steps in. He also set up custom skills, which are saved instructions for the AI:

• One skill copies his writing voice from 65 of his published articles, so documentation sounds like his blog.
• Other skills push the AI to follow existing code habits, since maintenance takes far more time than writing new code.
• Heavy models handle deep thinking. Lighter, cheaper ones handle release notes, reports and documentation checks.

A big migration with proof

Another Champion, who also coaches teams, moved about 900 teaching slides across 15 presentations from Reveal.js to Marp, two tools for building slide decks. Many slides had code, animations and live examples. By hand, he says, it would have taken a week or more, and he might have skipped it.

He and Copilot first analyzed the old material, set a target, wrote a plan and listed what had to stay intact. Then came checks. They compared screenshots of both versions, built a small app to run the decks side by side, and compared the PDFs. As he put it, the agent did not just say the migration was complete. There was evidence. The job took roughly eight to ten hours.

What to do

Anyone who pays for custom software can ask the development team two questions. Do you use AI coding tools? And who checks the code before it reaches our business? These Champions answer the second one the same way: a person does.

What is not known

The article text provided is cut off, so the third Champion's examples and tips are not covered here. The material also gives no figures on overall quality or bugs, only the developers' own accounts of time saved.