Short memory and long memory on the AI Agents track. Short memory is the current thread: user goal, steps so far, tool results. Long memory is facts saved across sessions, such as a preference or an account id. Mixing them causes the agent to treat a one-off instruction as a permanent rule.
This lesson assumes you already worked through Tools are the action space.
The idea in practice
Keep short memory inside the run and delete it when the task ends. Write long memory only through an explicit save tool, and show the user what was saved. Retrieve long memory with a query, not by pasting the entire history into every prompt.
A concrete check
goal = {
'track': 'AI Agents',
'lesson': 'Short memory and long memory',
}
checks = [
'input available at decision time',
'score matches the real decision',
'failure case written down',
]
print(goal['lesson'])
for item in checks:
print('-', item)
Run the sketch locally if you have Python. The printout is a reminder of the checks, not a trained model. Replace the strings with the real inputs from your own example before you treat it as a design.
What usually goes wrong
Saving every utterance as memory makes later runs obey outdated or private text. When this happens, stop adding parameters or tools. Fix the check, the data, or the permission, then run the same example again.
What to write down
- The input you are allowed to use at decision time.
- The output and the score or pass rule.
- One failure you will test on purpose.
- What you will not claim the system can do.
Practice
Split ten sample messages into 'this run only' and 'save for next time'. Explain one you would refuse to save.
Self-check
- Say Short memory and long memory in one sentence that mentions an input and an output.
- Name the failure mode in this lesson and the check that would catch it.
Done when: you can explain this lesson without the page open, and you have a written failure case.