Agent Interaction & Memory

Question 1

What is the human analogy used to explain agent delegation?

Answer 1

A boss delegating a task to an intern or expert.

Question 2

What is the simplest form of agent-to-agent coordination described?

Answer 2

Message passing.

Question 3

In message passing, what is the “message” from the boss to the intern?

Answer 3

The task description (requirements) sent to the intern.

Question 4

In message passing, what is the “message” from the intern back to the boss?

Answer 4

The result of the work (e.g., a report section).

Question 5

Why is message passing considered simple?

Answer 5

It’s just “send task → do work → return result.”

Question 6

What is the biggest dependency in message passing?

Answer 6

The quality of the task description.

Question 7

Why does task description quality matter so much?

Answer 7

The delegate’s output quality is directly dependent on the clarity and completeness of the task description.

Question 8

What common mistake can happen when tasks are sent via tool calls (e.g., JSON fields)?

Answer 8

The sender writes overly short task descriptions because the field looks “small.”

Question 9

Why can short task descriptions be harmful?

Answer 9

They omit context and requirements the delegate needs to do a good job.

Question 10

In multi-agent systems, what does the calling agent need to be good at?

Answer 10

Writing rich task descriptions for handoffs.

Question 11

What does a call_agent or handoff tool do conceptually?

Answer 11

It sends a task message to another agent and receives a result.

Question 12

Why might a coordinator not know what context an expert needs?

Answer 12

The coordinator lacks deep domain knowledge about the expert’s work.

Question 13

What does this create for task-based delegation?

Answer 13

A context-selection challenge: what information should be included in the message?

Question 14

What is an alternative to carefully crafting a rich task message?

Answer 14

Hand off the entire work log/memory so the expert can choose relevant context.

Question 15

What is the “work stream” in the analogy?

Answer 15

The ongoing log of work, inputs, and drafts collected so far.

Question 16

What does “handoff the entire log” enable?

Answer 16

The expert can read the history and decide what information is relevant.

Question 17

In agent systems, what plays the role of the “work log”?

Answer 17

Memory (conversation history + tool actions + results).

Question 18

What is “memory handoff” in this lesson?

Answer 18

Passing the agent’s memory to another agent so it can use full context.

Question 19

What is a key benefit of memory handoff?

Answer 19

The specialist can select the context it needs without the coordinator guessing.

Question 20

What is the simplest memory-handoff return style described?

Answer 20

The invoked agent reads the memory and adds one final result item back.

Question 21

Why might you want more than just the final result from the specialist?

Answer 21

To inspect intermediate work and reuse useful discoveries.

Question 22

What is a “fully shared memory” model?

Answer 22

The delegate sees the full memory, adds intermediate steps to it, and returns the complete updated log.

Question 23

What does fully shared memory allow the boss/coordinator to see?

Answer 23

Not just the final answer, but the delegate’s process and intermediate findings.

Question 24

What is a key upside of fully shared memory?

Answer 24

Transparency into the delegate’s work and richer collaboration.

Question 25

What is a key downside of handing off the whole memory to a specialist?

Answer 25

It exposes the specialist to irrelevant past context, increasing complexity and distraction.

Question 26

How can memory handoff reintroduce the problem multi-agent systems were meant to solve?

Answer 26

It increases each agent’s context load instead of keeping agents isolated and focused.

Question 27

What is the main tradeoff between message passing and memory handoff?

Answer 27

Message passing requires great task descriptions; memory handoff reduces that need but increases context complexity.

Question 28

Why might merging a specialist’s entire memory back into the coordinator be problematic?

Answer 28

It may flood the coordinator with details it doesn’t need.

Question 29

What is an example of “details the boss doesn’t need”?

Answer 29

The full list of calendar availability checks; the boss only needs whether scheduling succeeded.

Question 30

When might a specialist actually need full shared memory?

Answer 30

When it needs many details from the ongoing workstream (participants, constraints, emails, timing).

Question 31

What is the core design decision this lesson highlights?

Answer 31

How agents coordinate: task message vs memory handoff vs fully shared memory.

Question 32

What does the coordination choice affect most?

Answer 32

What context the delegate sees and what context comes back to the coordinator.

Question 33

Why does context flow “dramatically affect” reasoning and prompting?

Answer 33

More/less context changes what the model can infer, what it focuses on, and how reliably it follows requirements.

Question 34

What does “context selection” mean in agent interaction?

Answer 34

Deciding which information is passed to the other agent (or letting it choose).

Question 35

What is the main risk of giving too little context?

Answer 35

The delegate can’t do good work due to missing requirements.

Question 36

What is the main risk of giving too much context?

Answer 36

The delegate gets distracted, costs increase, and irrelevant info can pollute decisions.

Question 37

What is a practical hybrid approach implied by the lesson?

Answer 37

Pass a task message plus a curated subset of memory/context rather than everything.

Question 38

What is the key “engineering mindset” takeaway?

Answer 38

Delegation isn’t magic—communication format determines performance.

Question 39

What is the “intern” lesson for agent builders?

Answer 39

Clear delegation requires either good instructions or access to the right context history.

Question 40

What should be true of a handoff task description in message passing?

Answer 40

It should be rich enough to include requirements, constraints, and needed context.

Question 41

What should be true of memory sharing if you choose it?

Answer 41

It should be intentionally scoped so agents aren’t buried in irrelevant history.

Question 42

What is the central theme of “Agent Interaction & Memory”?

Answer 42

Agent collaboration depends on how information is shared, and that choice has major performance consequences.