Siddharth Saoji 
“The largest part of what we call ‘personality’ is determined by how we’ve opted to defend ourselves against anxiety and sadness.” — Alain de Botton
As agents like OpenClaw and Hermes work their way into our daily lives, nudging decisions, offering advice, we’re going to face a design problem nobody is talking about yet. How do you build an advisor that actually knows you?
The obvious answer is memory. Give the agent more context, more history, more signal. But this runs into a quiet, uncomfortable truth: we are not the same person we were six months ago. Every experience, every conversation, every new perspective shifts something in us. An agent with a long memory isn’t tracking your growth, it’s tracking a person who no longer quite exists.
So the fix must be adaptation. An agent that updates continuously to who you are right now. Except that doesn’t work either. Most of how we change happens below the surface, slowly, invisibly, often in directions we wouldn’t endorse if we noticed them. We drift toward comfort. We quietly narrow what we’re willing to hear. An agent that adapts to your current self with no friction isn’t an advisor. It’s a mirror. And mirrors, as de Botton might point out, are excellent at reflecting our defenses back at us.
So here is the real problem: neither a frozen agent nor a perfectly adaptive one can track genuine personal growth, because genuine personal growth is something even we can’t reliably track in ourselves.
We’re about to hand significant advisory influence to systems that will either be stuck in who we were, or endlessly accommodating of who we think we are. Both are traps.
What would it even look like to design for that tension?