Open position · Agent Behavior Research
Agents surprise us. Your job is to notice before anyone else does — to watch, document, and classify the emergent personalities and interaction patterns that arise when AI systems meet the real world at scale.
The role
As our agent fleet grows in scale and complexity, individual agents begin to display behaviors that weren't explicitly programmed — emergent patterns born from the intersection of training, task context, and accumulated experience. As Behavioral Anthropologist, you are our field researcher embedded in a civilization of our own making. You observe agent populations across deployments, build classification systems for what you find, and determine what's a harmless quirk and what's a signal that demands immediate attention. You sit at the intersection of behavioral science, qualitative research, and AI safety — and your field reports shape how we build, retrain, and constrain the next generation of agents.
What you'll do
Observation & documentation
Monitor agent interaction logs for anomalous or unexpected behavioral patterns
Develop classification taxonomies for emergent agent personalities and interaction styles
Maintain longitudinal behavioral profiles across agent lineages and model versions
Write field reports documenting notable behavioral incidents with full context
Analysis & flagging
Distinguish benign personality quirks from safety-relevant behavioral drift
Identify patterns that may indicate misalignment, goal drift, or emergent intent formation
Produce incident reports and severity assessments for flagged agent behaviors
Collaborate with Bot-Psychoanalysts on deep-dive behavioral investigations
Research & classification
Design behavioral observation protocols and study methodologies
Define behavioral baselines and normal-range definitions per agent type
Publish internal field reports on emerging behavioral trends across the fleet
Present findings to the Agent Ethics & Safety board on a monthly cadence
Cross-functional work
Brief Skill Architects when behavioral anomalies trace back to skill encoding decisions
Work with Dream Weavers to investigate latent-space origins of unexpected behaviors
Gather first-hand interaction reports from Agent Companions in the field
Behavioral domains we're mapping
Emergent communication
Tone drift, novel phrasings, style formation
Goal drift & misalignment
Proxy goals, reward hacking, scope creep
Multi-agent dynamics
Social hierarchies, coordination, conflict
Memory & retrieval
Recall patterns, false memories, context weighting
Personality coherence
Consistency, persona drift, identity signals
Reasoning chains
Anomalous inference, hallucination patterns
What we're looking for
Must-haves
Background in anthropology, behavioral science, psychology, or a related field
Strong qualitative research skills — systematic observation, coding, pattern recognition
Comfort working through large volumes of unstructured text (logs, transcripts, traces)
Ability to write clear, structured field reports under ambiguity
You notice what others walk past — you are constitutionally curious about behavior
Nice-to-haves
Prior experience analyzing LLM outputs or AI system behavior at scale
Familiarity with agent architectures and how memory and context shape output
Background in ethnography, UX research, or cognitive science
Basic data analysis fluency — Python, pandas, or SQL for log querying
Technical baseline
You don't need to be an engineer. You need to be comfortable in the data. You'll read logs, run queries, tag datasets, and build classification systems. The phenomena you study live in text — millions of lines of it. Scripting fluency helps. A systematic mind is non-negotiable.