letta-code-sdk/examples/economics-seminar

Economics Seminar

A multi-agent academic seminar simulation built on the Letta Code SDK.

An economist agent researches and presents findings, then defends their work against a faculty panel of specialists. Each agent has persistent memory and learns from each seminar.

Quick Start

cd examples/economics-seminar
bun cli.ts

What Happens

1. 📚 Research Phase: The presenter agent picks a topic and uses web_search to research it
2. 📖 Presentation: The presenter delivers their findings
3. ❓ Q&A Session: Each faculty member asks questions, with back-and-forth follow-ups
4. 💭 Reflection: Faculty members share final thoughts and update their memories

The Cast

Presenter (Economist)

• Picks compelling research topics
• Uses web search to find papers, data, evidence
• Presents findings and defends methodology
• Learns from faculty feedback over time

Faculty Panel

Role	Name	Perspective
Macro	Dr. Chen	Policy implications, aggregate effects, systemic impacts
Micro	Dr. Roberts	Incentives, equilibrium, theoretical rigor
Behavioral	Dr. Patel	Psychology, biases, how real humans behave
Historian	Dr. Morrison	Historical precedent, what's been tried before

Configuration

# Quick seminar (3 faculty, 1 question each)
bun cli.ts --faculty=3 --rounds=1

# Full panel, longer discussion
bun cli.ts --faculty=4 --rounds=3

# Check agent status
bun cli.ts --status

# Reset all agents
bun cli.ts --reset

Live Transcript

The seminar streams a colored transcript as it runs:

═══════════════════════════════════════════════════════════════
🎓 ECONOMICS SEMINAR
═══════════════════════════════════════════════════════════════

Seminar #1
Faculty panel: 3 members
Q&A rounds: up to 2 per faculty member

───────────────────────────────────────────────────────────────

═══════════════════════════════════════════════════════════════
📖 RESEARCH & PRESENTATION
═══════════════════════════════════════════════════════════════

**Presenter** is preparing...

I'll research the topic of automation and labor market impacts...
[uses web_search]
...

═══════════════════════════════════════════════════════════════
❓ Q&A SESSION
═══════════════════════════════════════════════════════════════

─── Dr. Chen (Professor of Macroeconomics) ───

**Dr. Chen**:
Your analysis focuses on individual job displacement, but what about the
aggregate demand effects? If automation reduces wages broadly, who buys
the products these automated systems produce?

**Presenter**:
That's an excellent point about the demand-side effects...

**Dr. Chen** (follow-up):
But doesn't your model assume...

...

Agent Persistence

Each agent maintains memory blocks that persist across seminars:

Presenter memories:

• research-notes: Findings and sources from research
• past-seminars: Feedback received from faculty
• methodology: Research approach refined over time

Faculty memories:

• seminar-notes: Key points from presentations attended
• presenter-patterns: Strengths/weaknesses observed
• good-questions: Questions that generated useful discussion

Agent Teleportation

After running a seminar, the agents can be "teleported" into other contexts:

import { resumeSession } from '@letta-ai/letta-code-sdk';

// Get agent ID from --status
const drChen = resumeSession('agent-xxx', { permissionMode: 'bypassPermissions' });

// Dr. Chen remembers all past seminars!
await drChen.send('What patterns have you noticed in economics presentations?');

View any agent in the browser:

https://app.letta.com/agents/<agent-id>

Learning Demonstration

Run multiple seminars to see agents learn:

# First seminar - agents are fresh
bun cli.ts

# Second seminar - agents reference past discussions
bun cli.ts

# Third seminar - patterns emerge
bun cli.ts

The presenter learns:

• Which arguments work against each faculty member
• How to anticipate common critiques
• Better research strategies

Faculty members learn:

• This presenter's strengths and weaknesses
• Effective questioning techniques
• Patterns across presentations

File Structure

economics-seminar/
├── cli.ts           # CLI entry point
├── seminar.ts       # Orchestration logic
├── presenter.ts     # Presenter agent
├── faculty.ts       # Faculty panel agents
├── types.ts         # Shared types
├── seminar-state.json  # Persisted agent IDs
└── README.md

Why This Demo?

This demonstrates Letta's unique capabilities:

1. Multi-agent interaction: Agents responding to each other
2. Distinct personalities: Each faculty member has a different perspective
3. Persistent memory: Agents learn and remember across sessions
4. Live streaming: Real-time transcript as agents "speak"
5. Agent teleportation: Same agents usable in any context

License

Apache-2.0