Handing real work to AI agents
What happens when an everyday user gives an AI agent an actual job? Each experiment records the assignment, how long it took, how much usage it consumed, what came out, and where it fell short.
What each experiment will record
- The tool used, such as Grok Bot or Muse
- The assignment, in the words it was given
- Time required and usage consumed
- The outcome: worked, partly worked, or failed
- Limitations, plus an honest call on usefulness
How results are recorded
- Failed and half-finished runs get published too
- Time and usage are recorded, not estimated afterwards
- Opinions about usefulness are labeled as opinions
No experiments are published yet. The first write-ups will come from real runs.