← All speakers

Bio, Work & Ideas

Tony Fabrikant

Conference affiliation: Co-founder · CoupleWork AI · 2026

Tony Fabrikant is a co-founder of BetterLabs AI and CoupleWork, developing clinically informed relationship coaching for situations where a chatbot’s agreeable answers can cause real harm. Previously a chief technology officer at S&P Global, he spent more than two decades in financial technology before applying his experience with security, privacy, and sensitive information to relationship AI.

At BetterLabs AI, Fabrikant partnered with psychotherapist Clay Cockrell to create Maxine, CoupleWork’s voice-and-text relationship coach. The system draws on the Gottman Method, Emotionally Focused Therapy, and attachment theory while distinguishing coaching from licensed mental-health treatment. Its multi-agent architecture maintains separate private and shared conversation contexts, helping couples receive continuous support without exposing one partner’s confidential disclosures to the other.

  • Clinician-led agent evaluation: Fabrikant translates clinical expectations into extensive test-driven evaluations, repeatedly probes difficult scenarios, and investigates even isolated safety failures. He also insists engineers use their own products: automated checks can miss subtle deterioration in an agent’s emotional tone. His AI Engineer conference appearance frames clinical expertise as an engineering input, not a final review.
  • Safety escalation and bounded coaching: CoupleWork is designed to recognize signs of coercion, abuse, self-harm, and acute distress, then direct users toward appropriate human support when relationship coaching becomes unsafe.
  • Privacy-aware relationship AI: Fabrikant applies strict boundaries between individual and joint sessions and encryption to conversations containing intensely personal relationship information.
  • Level of Attention: In his writing on AI-assisted development, Fabrikant argues that software delivery is increasingly constrained by qualified human judgment, not code-generation effort. Changes affecting model behavior, privacy, or safety require more expert scrutiny than larger, low-risk interface updates.

Read the topics behind these talks

1 conference talk

References