Honestly I think we're going to see a lot less AI-written code in the future, and more AI-assistance (PR reviews, documentation, security scans, scaffolding, brainstorming, test suites, etc). Example: to ship one feature for my open source project with a collaborator, we went back and forth for a month to agree on the change, test it, approve it, merge it. The code was pretty tiny. We could get more contributions faster if AI can help us tighten up that lifecycle.
Adding a new filter option for in-product analytics (assuming that data is being captured) is something AI can do reasonably well.
But things like “how should our orchestration layer be architected” isn’t a question of execution, so AI won’t be much use.