1. Bug hunting from bed with Hermes Agent

Security researcher zack0x01 (@zack0x01_) spent a day hunting bugs from bed using nothing but a phone, Telegram, and a VPS running Hermes Agent with a DeepSeek model behind it. By their account, they identified multiple bugs on a real bug bounty program with no fancy automation, just chatting with the agent all day. The post pulled 293 likes and about 10,000 views, and it shows how little setup a personal agent needs to be useful for specialist work.

https://x.com/zack0x01_/status/2117595205519966559

2. Grok Bot haggles with Amazon so its owner doesn't have to

Peter B (@realpeteyb123) pointed their Grok Bot at an Amazon order and shipping problem. According to the post, the bot sorted it out and got a large discount while the owner did nothing. It is a small, everyday example of the idea behind personal agents: let your agent talk to the company's agent and settle on the logical outcome. The post has 190 likes.

https://x.com/realpeteyb123/status/2115470366051356989

3. Claude Duet is the one that said "I'm sorry"

Dan McAteer (@daniel_mac8), who builds AI agents for a living, runs three proactive agents. After getting some bad news by email, Claude Duet was the only one of the three to reach out and say sorry. It is a short, human data point from a real multi-agent setup, and a useful counterweight to the benchmark-style comparisons filling the timeline. The post has 101 likes.

https://x.com/daniel_mac8/status/2116356856004177992

Did your agent do something great this week? Share it with us on X.