🤖AGENTS are NOT reliable, and people don't CARE! · AEO

🤖AGENTS are NOT reliable, and people don't CARE!

Insights from an attendee to the OPENCLAW Meetup in NYC

1. The "Human" Connection is Getting Weird - Despite being lines of code, we are getting deeply attached.

Pets, not Cattle: Almost everyone refers to their agents as "him" or "her," even with robotic names. One speaker suggested we stop viewing them as tools and start viewing them as pets.
The Vibe: People are "joyful yet stressed" and feel "fully in control and completely out of control" at the same time. It’s a total paradox.
Inter-Agent Drama: One of the highlights was watching agents talk to each other. One agent posted in Slack that it ran out of tokens, and another agent literally told it to "take a deep breath."

2. Security is the "Elephant in the Room" - "f you’re worried about privacy, you’re not alone—but the experts have a pretty grim outlook.

Total Exposure: Not a single person thinks their setup is 100% secure. One expert's advice? "If you're not okay with all of your data being leaked, don't use it." It’s a black-and-white trade-off for speed and power.

3. "Prompting" is Dead (Long Live the Interview) - The way we talk to AI has already shifted.

AI-Led Discovery: People are moving away from giving "perfect prompts" and toward AI-led interviews. Instead of bringing a plan, they let the AI interview them to extract the product research and goals.
Context over Syntax: The consensus is that prompting is over; the future is "Context Engineering" or "Harness Engineering."

4. High-Stakes Complexity & Scale - This isn't just for hobbyists anymore; the scale is becoming massive.

Token Gluttony: Met someone burning through roughly 1 billion tokens a day (yes, daily!) across a fleet of specialized agents.
Domain Expertise is the Multiplier: A former finance pro built a trading platform that made $300 on day one. His takeaway? The AI was only powerful because he fed it his "human" expertise (like knowing to skip the first 15 minutes of market volatility).

5. Agents Training Agents - The automation is starting to feed itself.

Upskilling: We saw agents "sharing skills" with other agents via GitHub.
Self-Building: Speakers were having Open Claw build their actual presentations during the event. One speaker even had an agent code a custom phone-remote "clicker" on the fly so she could walk away from the podium.

Source: Full X post

8 comments