Clear ideas for complex systems
Latest posts
Glossary

Long-Horizon AI Agents

Long-horizon AI agents perform tasks over many reasoning and tool-use steps, so early mistakes can alter later context, state and decisions.

Example

An agent that misreads a requirement, writes the wrong file and then uses that file as evidence can continue producing plausible output while the task has already drifted from its goal.

See Why Long-Horizon AI Agents Fail: The Mathematics of Compounding Errors for the full analysis.