Description
Mission
Salesforce is shifting from intent-based chatbots to autonomous, multimodal agents that plan, decide, and act across clouds. As a Lead Agent Behavior Designer, you design how these agents behave over time. You shape how they reason, take initiative, use tools, ground their actions, and recover from errors in text, voice, and visual surfaces. You turn product goals into reliable agent behavior that works for real users in complex enterprise environments.
You influence teams through craft, clarity, and deep multimodal expertise. You create behavioral patterns, voice strategies, and reusable structures that help engineering and PM teams ship consistent, trustworthy agents.
Core Responsibilities
(Lead-level: pattern ownership, cross-team guidance, forward-deployed impact, but not org-wide governance)
Agent Behavior and Multimodal Action Design
- Design how agents plan, act, confirm, and recover across text, voice, and UI surfaces
- Define behavior signatures for initiative, tool use, grounding, and decision quality.
- Shape visual and voice cues that help users follow agent reasoning and task progress.
Voice Agent Design and Turn Management
- Design how voice agents listen, speak, interrupt, and confirm across long tasks.
- Write natural speech patterns that support pacing, grounding, and user confidence.
- Model edge cases: silence, barge-in, latency, and state loss.
Reasoning, Memory, and Context Handling
- Partner with engineering and Ontology to define how the agent uses memory, evidence, and retrieved context.
- Identify gaps in state tracking, grounding, and contextual drift.
- Propose fixes based on transcripts, logs, and failure signatures.
Tool Use and Orchestration Patterns
- Design behaviors for when agents choose tools, how they report actions, and when they ask for user direction.
- Document patterns that clarify tool boundaries, confirmations, and rollback paths.
- Prototype sequences that show how multimodal actions support user understanding.
Prompt, Instructional Logic, and Behavior Rules
- Write prompts and behavioral instructions that shape reasoning, tone, and decision steps.
- Tune prompts for text, voice, and multimodal scaffolding.
- Translate ambiguous product goals into clear behavioral instructions for engineering teams.
Reusable Agent Patterns and Behavior Libraries
- Create templates for confirmations, grounding, multimodal transitions, tool sequences, and recovery moves.
- Document patterns in a format that partners can adopt quickly.
- Validate new patterns through delivery work, customer pilots, or controlled testing.
Agent Evaluation and Drift Detection
- Apply agent heuristics to measure clarity, context use, respectfulness, and action quality.
- Identify drift in reasoning, tone, or tool choice across modalities.
- Share clear findings and propose changes grounded in observable behavior.
Cross-Cloud Delivery and Partnership
- Lead design for projects that require strong partnership with engineering, PM, and Trust.
- Highlight where behavior, memory, or tool gaps affect user outcomes.
- Help teams adopt agentic patterns that meet quality, safety, and tone standards.
Mentorship and Skill Development
- Guide designers in multimodal and agent behavior craft.
- Model best practices for testing, failure-mode analysis, and pattern reuse.
- Share examples, reviews, and artifacts that raise quality across teams.
What You Bring
- Deep experience designing for agentic systems, reasoning behavior, and multi-turn workflows.
- Strong writing for both voice and text.
- Skill in shaping agent behavior across modality shifts.
- Knowledge of memory models, plan-and-act loops, and RAG-driven reasoning.
- Ability to diagnose agent failures through logs and behavioral signatures.
- Experience creating reusable behavioral patterns that scale.
For roles in San Francisco and Los Angeles: Pursuant to the San Francisco Fair Chance Ordinance and the Los Angeles Fair Chance Initiative for Hiring, Salesforce will consider for employment qualified applicants with arrest and conviction records.
