
Everyone is building AI agents. Michael Waitze asked the hard questions: "What makes Maibel different - and HOW do you actually build it?”
His immediate verdict: what we're building sounds incredibly hard. He was right.
Most agents optimise for completing a bounded task, but Maibel must maintain coherent judgment with each woman across months.
Each user experiences one consistent companion. Behind that experience, our multi-agent orchestration coordinates isolated memory and state, behavioural decisions and safety.
The hardest part isn’t storing context. It's resolving conflicting considerations and testing whether the resulting behaviour remains coherent across multi-turn trajectories - not merely whether ONE single response looked accurate or empathetic.
Thank you Michael Waitze and Asia Tech Podcast for pushing into the real technical difficulty behind persistent companion AI. There's no settled playbook for doing this well and that's precisely why we’re building it.


