pencil - Agent Architect

• 3+ Years of Direct GenAI Experience: You have a deep, intuitive, and technical understanding of LLMs (GPT-4, Claude, Gemini) and multimodal models (Stable Diffusion, Midjourney, Video Gen). • Systems Thinking: You don’t just write a prompt; you think about the latent space, the context window, and how one agent's output becomes another’s input. • Technical Literacy: While this isn't a "Software Engineer" role, you should be comfortable with Python, JSON structures, and API documentation. Experience with agent orchestration patterns (task decomposition, tool use, state management) matters more than any specific framework. • Evaluation Obsession: You believe that if you can’t measure a prompt’s performance, you shouldn’t ship it. You are familiar with benchmarking and A/B testing AI outputs. • The "Creative/Technical" Bridge: You can sit in a room with a Creative Director and translate "make it feel more punchy" into a concrete change to the system. You are comfortable being the face of the technology in client meetings, not just the person behind it. • Client-Facing Problem Decomposition: Demonstrated experience leading discovery with customers or stakeholders and turning vague briefs into scoped technical solutions. Solutions architecture, forward-deployed engineering, or technical consulting backgrounds fit well here. • Plain-Language Communication: You can explain agent failure modes, eval results, and architectural trade-offs to a CMO without jargon, and you enjoy doing it. • You’ll Thrive Here If... • You find "hallucinations" to be a logic puzzle to be solved, not just a bug. • You are excited by the challenge of making an AI follow a 50-page brand book with 100% fidelity. • You want to build the infrastructure that defines how the next generation of advertising is created. • You’d rather sit with a confused client and untangle their problem than receive a perfect spec. • KPI & Success Measures • System Reliability: Reducing the "failure rate" of complex agent workflows. • Architectural Efficiency: Minimizing token usage/latency while increasing output quality. • Brand Alignment: Scoring of agent outputs against specific brand guidelines using automated evaluation tools. • Adoption at Scale: Core agents adopted and retained across the Pencil user base, not just shipped to it. • Client Deployment Success: Custom client workflows delivered on time and adopted by the client, measured from brief to live, not build complete. • Stakeholder Trust: Clients and internal partners come back to you to scope the next problem. Repeat engagement on strategic accounts and positive feedback on clarity of communication.