You begin with an present challenge and the main points of construct instruments and frameworks, together with a goal Java model (for instance upgrading from Java 8 to Java 21). The Copilot improve assistant analyses your code base and generates an inventory of the required steps to run your improve, presenting it as a set of GitHub points you could test earlier than operating the replace.
When you’re pleased with the duties, the instrument takes you to a dashboard the place you possibly can watch the replace course of, together with how Copilot rewrites code for you. You’ll be able to cease and begin the method at any time, drilling down into duties for extra info on simply how the AI-based code is working. It’s good to have this degree of transparency, as you want to have the ability to belief the AI, particularly when it’s engaged on business-critical software program.
As that is an agentic AI course of, the service can detect errors and repair them, launching sub-agents that make modifications, rebuild, and retest code. Curiously if a repair doesn’t work, it’ll take one other method, utilizing the shared information of the Java builders whose work has been used to coach the Copilot Java mannequin. Like different GitHub Copilots, modifications that work are used to fine-tune the mannequin, lowering the danger of errors in future runs. That goes for guide updates and modifications too.
