Agents talking to agents

My agent submits a new feature for review.

One of our engineers responds “My agent made some suggestions, see here <file>”.

I tell my agent to address the suggestions by the other agent.

Yet another agent does an independent review.

My agent also addresses those, and replies to questions.

The (human) engineers merges the new feature.

This all happened this week. Agents and humans working together, but humans mostly acting as glue. And if you squint - the glue isn’t even that complicated to make (in my case, I run an agent in a loop and tell it to check for comments every 30 minutes).

Subsequently, another interesting thing happened, which is that my agent started to reply to comments of my colleagues. But the agent was using my account, so it seemed like I was replying to them. And in turn, they replied directly to my comment, which in reality was written by my agent.

So what I did was I told my agent that going forward, if you post using my account, please say that you are my AI instead of replying as if you were me without any further notes.

It made me realize that there has to be - and yes, I realize I’m not the first person to realize this - there has to be some sort of process or standard by which we say this is an agent doing something. Alternatively, this is an agent acting on behalf of somebody, or this is a human writing the words very directly.
If only to prevent the situation in which somebody says, “Ah, I saw your comment,” to which you have to reply, “What comment?”.

Regardless, we live in an incredibly exciting time. It’s really fun to see new developments happen almost daily and see how they meaningfully can impact our work and our life, arguably very positively.