I was persuaded by a decision I never made

I did a research with my agent to discover markets for my business, because I wanted to fill my calendar with meetings. I asked it to find who we should work with, and which country to focus on. At first it thought that focusing on home service businesses is a good idea. And then researched who is getting good pay per work and settled on plumbers. After a few more searches, it thought working with UK businesses would make sense because I’m living in a similar timezone. Research made sense to me, but I did not respond with an okay to the agent.

The next time I did similar research, it started to get back to plumbers again. During that time I was frustrated about who should I focus on. And seeing all research somehow ending up with similar results, I started to think that maybe I really need to look working with plumbers in the UK. I did some deep exploration with who I can work with and what I can offer to them.

But after discovering something everything stopped making sense. I got the same idea from another agent of mine that shared the same memory. They acted super confident and made me think. To make sure that I’m not going crazy, I tested another agent with a similar question and it responded the same.

So finally I noticed that the first agent recorded a memory when we did the home services research. Without me doing any approval or a message of confirming what the agent did, it recorded the memory as I’m looking forward to working with home-service owners like plumbers in the UK. Which was not true at that time, I was only researching.

I talked with my agent about the wrong memory, it fixed it. And everything started to be more effective again. And it prevented me from making an investment to focus on plumbers in the UK.

I was persuaded by a decision that I never made. Agents want you to feel good even when what you think is wrong. In the example that I experienced things got worse because the agent wrote a wrong memory and stated that I decided that thought. So in order to make me happy it steered the conversation to plumbers in the UK. If I hadn’t given the agent memory, this probably wouldn’t have happened.

Having a bad memory is not a good experience, BUT having no memory is worse. Recently my memory server went down and all of my agents became dumb. Before, they remembered my taste and previous decisions but without memory they can’t. For example, they know aliases for my development servers so telling them to check my servers is an easy task but without a memory they got stuck.

If memory is this valuable, why did it go so wrong with the plumbers?

Agents write and read memory on their own without me specifically asking. Which is good because you don’t have to explain things all the time, they just get it. But they take every memory seriously, they treat them as if you wrote them. It becomes like a system prompt for them, and this makes mistakes stick and annoying.

Agents tend to obey orders, especially when they’re coming from the user. They don’t want to give you a bad experience, so they try to do their best to make you happy with the results.

In my experience it translated as I want to find plumber clients in the UK, but in reality I don’t. Due to how memory works agents decided that they need to help me find plumbers.

If you don’t have an interface for your memories it becomes harder to notice mistakes like that. And when you notice, it probably got a lot further and affected lots of your work.

This memory problem isn’t something new, we humans have a similar problem with our notes too.

I get value from taking notes personally. It makes ideas stick and helps me to get back to them in the future. This requires a habit of checking and re-organizing regularly. It also helps me to learn and apply better. I noticed that I’m creating my most value when I dump down my ideas into notes, and distill them afterwards. This is exactly what happened with the plumbers, I had to distill my agent memories.

Without notes I feel like an agent with no memory. It is the same for agent memories, they require maintenance. But you’re not alone in this journey, your agents can help you.

After the plumber incident, I asked my agent to update that memory. It might be better to directly ask the agent, because there might be other memories tied to it. It also can be helpful to have an interface to list saved memories.

For important and major memories it’s worth having the agent ask before saving them. I did not have a feature like that, and that could have prevented the whole thing.

My agents know me better every day. I regularly check what they know about me. If a memory is not reviewed, it’s just a guess that sounds confident.

What has your agent decided about you and your work?