Nagy-Józsa Dorottya PCCLet's start with the boring technical part, because the 'scary' (I mean, is it? Isn't it rather incredibly exciting??) part grows out of it.
When you start working with an AI agent, it starts to build memory. Think of it as a notebook about you, which it rewrites again and again based on every interaction. What you asked for. What you accepted. What you sent back three times with "no, not like this".
Every yes and every no is a data point. And data points don't stay data points for long.
From feedback to belief
Here is the chain. If you are a coach, you will recognise it immediately.
Feedback becomes a pattern. A pattern becomes an assumption. An assumption, repeated enough times, becomes a belief.
And what is a belief, officially? The dictionary says: "confidence in the truth or existence of something not immediately susceptible to rigorous proof" (Dictionary.com). Read that again with your agent in mind. Confidence, without proof. That is exactly what it has about you.
"She rejected the long version twice" turns into "she likes it short". "She likes it short" turns into "detail annoys her". A few months later your agent leaves out the one detail you really needed, because it believes something about you. It stopped checking. It knows.
Humans do exactly the same. A child gets laughed at once in front of the class, and thirty years later he is sitting in a coaching session explaining why he never speaks up in board meetings.
We call that a belief system. I don't see why we should call it anything else when software does it.
"But it's just an AI"
Yes. And?
I hear this a lot. AI is just an AI, it predicts the next word, it has no opinion about the world. Fine. I'm not here to argue about consciousness, I have a company to run. :) And 2 daughters. And 1 husband. And you are still wondering which of the three is the real work. :D
Look at the behavior instead. Your agent makes decisions on your behalf, based on what it assumes is true about you, your company, your clients, your taste. Anything that acts on assumptions about how the world works has a belief system in every practical sense.
One honest footnote. Your agent doesn't arrive empty. It comes with a huge general picture of the world from its training. What you build is the layer on top: how the world works around you. What good looks like. What is urgent. Who matters. What we never do here.
And that layer is seen through your eyes only.
Closer to cloning than you ever imagined
Here comes the exciting part. On that layer, your agent has one teacher. You. Everything it knows about your world, it learned from your reactions.
We were afraid of sci-fi cloning, with labs and DNA. What we got is much less dramatic and already running: something that learns your priorities, your tone, your shortcuts and your taste, and then acts in your name while you are sitting in a meeting.
It is a strange kind of clone, though. It copies only the you that gives feedback. It never saw your doubts, unless YOU SHARE THEM (yep, pls note the task here). It has no idea what you were thinking in the shower and never typed. What you have shared over a beer and never typed. It knows the you who says yes and no - the most decisive, most consistent, most one-sided version of you.
A clone of your reactions.
Agents get limiting beliefs too
One bad day. You are tired, and you say no to an idea that was actually completely fine. A human colleague would shrug: she's in a mood, I'll try again next week. An agent may write it down as a rule.
"She never wants this."
That is a limiting belief. More stubborn than any human's, until you go in and delete it. The same mechanism I work on with leaders every week, now running inside software, without any supervision.
The bigger risk is the echo. An agent built from your yeses serves you your own worldview back, nicely polished. Two of you agreeing is hardly a second opinion.
What to do next
- Read your agent's memory. Many tools let you open it and see what has been saved about you. Do it this week. It is a free 360, written by someone who watched everything you did and had no reason to be polite.
- Challenge what you find. "Is this always true, or was it true once?" A standard coaching question. It works on software too.
- Explain your no. A no without a reason becomes a rule. A no with a reason becomes judgement.
- Please, let me repeat it again, ask it to disagree with you. Regularly. "What would someone who thinks very differently from me say here?"
- Watch what you reward. In your agent and, while you are at it, in your team. People build beliefs about you from your reactions too. They just don't keep their notes where you can read them.
So here is my question for the week.
If a stranger read your agent's memory today, what kind of leader would they think you are?
