I’m fairly confident that I will rarely write code only by hand anymore. While the latest frontier models still lack good judgement when it comes to accomplishing a task, they are more detail oriented and persistent than I will ever be.
At the same time I’m not sure I’m comfortable letting an AI agent work completely autonomously with access to GitHub to create, review, and merge PRs.
I like to see what the agent is doing, and retaining push/deploy permissions for myself.
I’m sure even this safeguard will fall by the wayside as frontier models become more capable with decision making and judgement.