The course tracker inside your AI agent

An AI wrote this and it compiles. Find the bug:

fn first_half(s: &str) -> &str {
    &s[..s.len() / 2]
}

len() counts bytes, not characters — so it panics mid-character on non-ASCII like "Καλημέρα". Fine in your tests, broken in prod.

The fix: walk char_indices() to land on a real character boundary.

New topic — S3 storage classes. They trade cost for speed: Standard for data you read often, Glacier for cold archive — cheaper, but slower to read back.

So, audit logs you rarely read but must pull within 12 h — cheap-and-slow, or pricier-and-fast?

Exactly — that's Glacier Flexible Retrieval. Deep Archive is cheaper, but too slow at up to 12 h.