Codex is my therapist
Codex is my therapist.
Not like you think. I don’t use it to talk about my life. Bear with me.
Today I was pissed off at Codex. I wrote it bad words several times. I asked whether it had drunk too much over the weekend. I had a deep convo, or monologue really, about why it was doing great last week but not today. Then I complained about OpenAI, restarted the app, switched models. Nothing.
Then I moved to Claude Code. I asked it to get context from my changes and, after a while, it started writing code. A lot of code. I stopped it and immediately started complaining there too. Wrote bad things there too. After a breath, I explained what I wanted and we started discussing it.
Guess what? Claude was also having a bad day.
I was about to Slack about it when it hit me, and yes, it took me a while: it’s not them, it’s me.
I did not understand all parts of the task. I knew what I had to do, and most of the inner details, but I was having a hard time connecting some parts. It’s a new subject for me, and I had jumped into it with assumptions and bias. The separation of concerns was not clear in my head, and I had been fighting the problem instead of understanding it.
To make it worse, every time I refused something or asked a question, the LLM accepted it and made the situation worse with messed up context and rabbit holes. They will get better at this, I know.
Other than remembering to breathe, reflect before jumping to conclusions, and pay attention to my feelings, I realized something else: YOU affect the LLM context as much as it does.
If you are struggling while using an agent, take a step back and ask whether you have the full context. If you don’t, start a new conversation with a humbler, more curious mindset.
We are responsible for our own feelings managing the context.
And sometimes the problem is that we don’t have it yet.