Research & writing

What we're learning while building Stoa, an AI relationship guide that gives honest perspectives instead of just agreeing with you.

• September 2026

  Stoa is 36% less agreeable than ChatGPT

  ChatGPT sided with the person in the wrong 77% of the time. Stoa: 49%. We measured how often AI assistants just agree with you, across 1,840 conversations.

• Research · September 2026

  Measuring agreeableness in AI relationship advice

  The full research report: 1,840 conversations, 15 arms, methods and limitations.

• May 20, 2026

  How stoa handles conversations about suicide and self-harm

  The design philosophy and specific decisions behind how stoa handles conversations that surface self-harm or suicidality.