*if* you could rebuild collaborative software production, from first principles, how would you do it? start with good linting + formatting + unit tests red-green-refactor approach (slightly "adversarsial")? or allow dev to update tests as they go ("collaborative")? do you write evals for "good" unit tests, that can be judged by LLMs? are integration tests useful? regression tests? what if you have an agentic browser that can cycle through all user journeys if you can waterfall the development, tests are really powerful. if you need to be more agile, how can you write good tests before you discover details of what you're trying to accomplish? personally, I find unit tests fairly useless. Smoke tests are more efficient
- 1 reply
- 0 recasts
- 0 reactions
BFC: read 12-36 books a year AFC: read 0 books a year (my experience only, as a never-used-X-before-FC-eroooor)
- 0 replies
- 0 recasts
- 2 reactions
i been saying since time that PRs and code review are outdated for the agentic world when was the last time you confirmed the contents of a package version update? accepting PRs should become as brain-less as upgrading your OS version (because there's automated + manual layers of safety in the process)
- 2 replies
- 0 recasts
- 4 reactions