Hey HN community, I built MergeAtlas - a better way to review PRs when bulk of the code is written by AI agents. It is a skill for your coding agent plus a small python script.MergeAtlas was built from a simple premise - When AI agents are writing the code, there's no need to review every single line of code. Instead it is better to understand the intent, important lines, and the blast...
> There's hardly any work you can think of which can't be done faster / beter with ai assistance
True, and someone needs to be the creative brain behind the decisions. AI can help you implement. When I say help, I mean literally help because one-shotting and vague prompts can get you only so far, usually with a lackluster result. While they are good at analyzing solutions, and finding out holes in one's thinking, ultimately, it is some creative actor that needs to understand the bigger picture to evaluate trade-offs, understand scope creeps, and spot overengineered. For now that actor is a human.
> If you have an anti-example, would like to hear.
In my personal experience, especially with greenfield projects, smarter models tend to overengineer the solutions. However, I haven't used Fable and Astra models, maybe they are better at creative tasks without overengineering.