Uploaded July 2026 | Updated September 2026, 2 weeks ago
What happens when a personal AI agent can understand you, act across your tools, and grow by rewriting its own code?
In this Observe 2026 session, Chi Wang, cofounder of AG2 and creator of AutoGen, introduces “My AI Stan” — a personal multi-agent system that runs on his own hardware, works across voice, chat, cursor, screen, Discord, code review, experiments, and overnight self-improvement loops.
Instead of giving a traditional talk, Chi has his AI Stan help present the session live. The agent explains its evolution from AutoGen to AG2 to personal multi-agent systems, walks through its core pillars of understand, act, and grow, and demos how it can edit at the cursor, send updates to Discord, coordinate across multiple owners, run research experiments, review code, and improve itself over time.
You’ll learn:
- How Chi Wang thinks about the evolution from AutoGen to AG2 to personal AI agents
- Why personal multi-agent systems may be the next form factor after agent frameworks
- What it means for an AI agent to understand, act, and grow
- How an agent can operate across voice, chat, cursor, screen, Discord, and code
- How AI agents can coordinate across devices, owners, and fleets
- What it looks like when agents review code, catch mistakes, and write rules back into files
- Why memory and correction loops matter for long-running personal agents
- How “realtime by day, rewriting itself by night” changes the role of AI assistants
Chapters:
0:00 Meet Chi Wang and the AI Stan experiment
0:38 Letting the agent present the talk
1:18 From AutoGen to AG2 to personal multi-agent systems
1:45 Sutando: a personal spirit in the digital world
2:58 Understand, act, grow
4:06 What the AI Stan can do today
4:39 One agent, multiple forms
5:57 How Sutando rewrites and reviews its own code
7:59 Fleet growth across devices and owners
9:15 Multi-owner AI agent collaboration
10:41 Agents reviewing code with no human intervention
12:25 Live demo: editing at the cursor
13:45 Sending updates to Discord
14:18 Lab Code: running experiments for research
14:36 Codex vs. Claude Code experiment
17:22 What all the use cases have in common
17:47 Voice, chat, cursor, and vision
18:10 Realtime by day, overnight by night
18:42 Memory that rewrites itself when corrected
19:12 Self-direction and closing the loop
20:08 How to start with AI Stan
Presented at Observe 2026.
#AIAgents #AG2 #AutoGen
🔗 Try Arize AX & Phoenix OSS: arize.com
🔔 Subscribe for weekly content on LLMs, agents, and evaluation: youtube.com/@arizeai?sub_confirmation=1
What happens when a personal AI agent can understand you, act across your tools, and grow by rewriting its own code?
In this Observe 2026 session, Chi Wang, cofounder of AG2 and creator of AutoGen, introduces “My AI Stan” — a personal multi-agent system that runs on his own hardware, works across voice, chat, cursor, screen, Discord, code review, experiments, and overnight self-improvement loops.
Instead of giving a traditional talk, Chi has his AI Stan help present the session live. The agent explains its evolution from AutoGen to AG2 to personal multi-agent systems, walks through its core pillars of understand, act, and grow, and demos how it can edit at the cursor, send updates to Discord, coordinate across multiple owners, run research experiments, review code, and improve itself over time.
You’ll learn:
- How Chi Wang thinks about the evolution from AutoGen to AG2 to personal AI agents
- Why personal multi-agent systems may be the next form factor after agent frameworks
- What it means for an AI agent to understand, act, and grow
- How an agent can operate across voice, chat, cursor, screen, Discord, and code
- How AI agents can coordinate across devices, owners, and fleets
- What it looks like when agents review code, catch mistakes, and write rules back into files
- Why memory and correction loops matter for long-running personal agents
- How “realtime by day, rewriting itself by night” changes the role of AI assistants
Chapters:
0:00 Meet Chi Wang and the AI Stan experiment
0:38 Letting the agent present the talk
1:18 From AutoGen to AG2 to personal multi-agent systems
1:45 Sutando: a personal spirit in the digital world
2:58 Understand, act, grow
4:06 What the AI Stan can do today
4:39 One agent, multiple forms
5:57 How Sutando rewrites and reviews its own code
7:59 Fleet growth across devices and owners
9:15 Multi-owner AI agent collaboration
10:41 Agents reviewing code with no human intervention
12:25 Live demo: editing at the cursor
13:45 Sending updates to Discord
14:18 Lab Code: running experiments for research
14:36 Codex vs. Claude Code experiment
17:22 What all the use cases have in common
17:47 Voice, chat, cursor, and vision
18:10 Realtime by day, overnight by night
18:42 Memory that rewrites itself when corrected
19:12 Self-direction and closing the loop
20:08 How to start with AI Stan
Presented at Observe 2026.
#AIAgents #AG2 #AutoGen
🔗 Try Arize AX & Phoenix OSS: arize.com
🔔 Subscribe for weekly content on LLMs, agents, and evaluation: youtube.com/@arizeai?sub_confirmation=1
