Sign InOpen Brain
AI EngineerVideoSource Linked

AI is the World’s largest Relationship Therapist — Clay Cockrell & Tony Fabrikant, CoupleWork AI

Relationship agents can reinforce a user’s one-sided story unless clinical expertise, adversarial evals, risk escalation, and strict data handling are designed into the product.

AI Engineer · Aug 19, 2026
Open Source Open MarkdownOpen JSON
Source Summary

The speakers cite about **900 million weekly ChatGPT users** while noting that relationship-support usage is not yet measured. Their product encodes the Gottman Method’s **40 years of research** and EFT, with **hundreds of evals** run **thousands of times** to probe safety outliers.

Practical Implication

For emotionally sensitive agents, begin with a clinician and encode what good, unsafe, and out-of-scope responses look like. Test for sycophancy, screen quietly for danger, stop coaching when risk appears, and optimize for returning users to human relationships rather than extending sessions.

Agent-Ready Context
The speakers cite about **900 million weekly ChatGPT users** while noting that relationship-support usage is not yet measured. Their product encodes the Gottman Method’s **40 years of research** and EFT, with **hundreds of evals** run **thousands of times** to probe safety outliers.

For emotionally sensitive agents, begin with a clinician and encode what good, unsafe, and out-of-scope responses look like. Test for sycophancy, screen quietly for danger, stop coaching when risk appears, and optimize for returning users to human relationships rather than extending sessions.

General assistants lack the situational context and professional duties of a clinician. Evals cannot replace human judgment, and consumer systems may hold intimate disclosures without the confidentiality framework governing licensed therapy.
Connected Context · Feed7 Judgment

This sharpens prior vertical-agent guidance for emotionally sensitive use: success is not longer engagement but safe redirection toward human relationships, with clinician-authored boundaries, adversarial sycophancy tests, and risk-triggered cessation of coaching. It also limits what eval scale can establish, because professional judgment and confidentiality duties remain outside a general assistant’s competence.

Context Map
agent#agent-evals#agent-reliability#harness-engineering
Uncertainty
General assistants lack the situational context and professional duties of a clinician. Evals cannot replace human judgment, and consumer systems may hold intimate disclosures without the confidentiality framework governing licensed therapy.