MediaClip

A pile of do-nots, then approval for what it should do on its own

November 4, 2026

0:45

I watched a prompt to the agent team that was just supposed to update a quote. It got padded with "do not send any invoices, do not publish, do not this, do not that."

Then I got "a big old report saying, yeah, we didn't do that. We didn't do that. We didn't do that."

I don't know if this is actually possible in agents, but we see it in humans. Throw instructions like that at me all the time and "I'm not going to have any confidence." The moment I'm unsure, I start asking you to approve the thing I'm supposed to be able to do on my own. And that's behavior I've seen in agents.

If yours keep asking permission, check how many do-nots they were handed.

Share thisPost on X(opens in a new tab)Share on LinkedIn(opens in a new tab)

The next step

That's the moment. Here is how you make it yours.

Build something realTake what you learned and build something concrete. valuefirstagent.com is where your build becomes a working agent.valuefirstagent.com(opens in a new tab)

And then, the rest of the path:

  1. Join the conversationainativehumans.org(opens in a new tab)
  2. Practice with peersvaluefirstcollective.com(opens in a new tab)
  3. Teach the next personthe Academy