For AI agents, the hard part may be defining ‘good’

This Reddit post says the author runs AI agents throughout the day. The author says the bottleneck is not the model itself, but their own work of defining what counts as a good result.

Key points

  • The author says they run AI agents throughout the day.
  • They say the main bottleneck is not the model.
  • They point to the difficulty of defining a good result.
  • Clear success rules can matter as much as model choice.
  • Unclear goals can lead to extra retries and higher token use.

Quick term guide

AI agents
AI agents are AI tools that can carry out steps toward a goal, not just answer once.
AI agent
An AI program that can inspect information and suggest what to do next.
bottleneck
A point where work gets stuck because one person or step cannot handle the volume, slowing down everything else.
token budget
The maximum number of text chunks (tokens) an AI can process or generate in one step — more tokens means higher cost.
budget
The maximum amount of tokens or money an AI is allowed to spend on a single task.
benchmark
A test used to compare speed, quality, or cost.
Matter
A smart home standard that helps devices from different brands work together.
retries
Attempts to run the same task again after it fails or gives the wrong format.
Read original