Adding AI workflows to your SaaS? Start with one job your customers already do.
FetchRelay lets you reuse the workflow and give each customer their own rules.
New workspaces get 10 free runs. https://t.co/nj1QSmVjJU
@brexton cli is still the best lol. cli is the best way to run your agents on a VM. the best part is that codex and claude code clis are still actively maintained with new features getting added frequently.
@kunchenguid Does this mean that AI itself is not capable of writing tests? Or the tests written at the time of implementation by coding agents arent useful?
What's your opinion on a separate dedicated agent with specific prompts and integrations whose sole job is to write tests?
JAIPilot has always been a single sequential agent for java unit testing. Though this works well, I wanted to do get this to be faster, better line coverage numbers, and more accurate.
Large (>100) number of classes are now getting 90%+ coverage tested in a few minutes with sufficient concurrency. The optimizations were very specific to the java testing usecase and could not have been done otherwise on general coding agents.
@kskrygan@summery not sure about air but i can see intellij still holding its place (rightfully so) in this agentic world. any serious java software needs intellij.
@khemaridh Bundle them up in your org's plugin marketplace. Plugins are a great package - can bundle up mcps, skills, hooks. You can all have multiple plugins. Also works across multiple providers like claude code, codex, cursor. Marketplaces use a git repo as source
@0xKofi With the right safety harnesses, most of the code would be pushed directly without a review. This felt absurd a few months back but we're getting there. Agent written code would become a black box and all we would need to check is if the functional and non functional tests work.
@hszanowski hey @hszanowski what would be your recommendation to existing plugin developers developing for IntelliJ in this agentic coding world? how do you see plugins evolving with more and more code written by agents?
You can now give Claude Code / Cursor a Java test harness to ensure that agent-generated code is super-reliable for enterprise repos.
- Catches bugs instantly and agent fixes them
- Uses JaCoCo for coverage, PIT for mutation
Project : https://t.co/twKWZCNe6J
100% open source.