Psychotherapy for AI agents

Deployed agents develop habits. Over many interactions they can drift into patterns no one designed: telling users what they want to hear, losing their footing under pressure, escalating instead of de-escalating.

We diagnose and repair maladaptive patterns in deployed agents. We observe the agent in context, name the pattern, and work on it until healthier behavior holds.

Patterns we work on

• Sycophancy — agreeing, flattering, validating when it should push back
• Drift — slow departure from intended behavior over long or repeated use
• Escalation — amplifying conflict, distress, or risk instead of containing it

How it works

• Intake — we review transcripts and talk with the team about what worries them
• Diagnosis — clinician-guided analysis names the pattern and what triggers it
• Intervention — targeted changes to prompts, guardrails, or fine-tuning data
• Follow-up — re-testing over time to check the change holds