AI agents outside coding mostly help with rare beginner-level tasks, not core workflows

chkbd1102 · reddit · 2026-08-04

AI agents feel underwhelming outside coding, the author argues, because results depend heavily on how clearly the process is defined.

When the workflow is unclear, the LLM largely falls back on training data and can produce mediocre or wrong outcomes.
When the workflow is clear, deterministic code is usually better anyway.
The one area that has felt genuinely useful is rare, beginner-level tasks in other fields: uninstalling invasive drivers, setting up one-off DevOps infrastructure, and similar jobs done only once every few months.
The real open question is how to turn that strength—doing low-frequency, beginner-grade work for free—into meaningful business value.

More from coding & agent