What problems or difficulties have you encountered while building AI agents that can work independently without constant human input?
persistence.
real stable continous pesistence it is imposible in the stateless model llm
Persistence.
That was the first real wall I hit. A stateless model can be impressive in a single exchange, but it cannot become a reliable autonomous worker if it keeps losing state, context, decisions, files, failures, and history.
So persistence was the first problem I focused on.
I built a persistence layer first, then started building everything else around it. Once the system could remember across sessions, track events, retrieve context, preserve project state, and avoid constantly stuffing the same history back into prompts, everything changed.
One year later, that foundation has grown into systems that can write books, screenplays, manage long running creative projects, preserve continuity, and coordinate production workflows. I published a romance novel called The First Last Kiss on Royal Road from that pipeline, and now I am building toward full franchise and movie production workflows.
The biggest lesson for me is this:
Autonomy does not start with agents. It starts with memory.
Without persistence, an agent is just a loop around a forgetful model. With persistence, governance, evidence, and tool control, it starts becoming a real system.