Asks for

The ad says“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 ad says“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.”
The ad says“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.”
The ad says“Agent Architecture: Design and implement multi-agent workflows, including task decomposition, state management, and tool-use (RAG, API integration, etc.).”
The ad says“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 ad says“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).”
The ad says“Systematic Optimization: Move beyond "vibe-based" testing. Implement rigorous evaluation frameworks (e.g., using LLM-as-a-judge, promptfoo, or DSPy) to measure and improve agent performance at scale.”
The ad says“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).”
The ad says“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).”
The ad says“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.”
The ad says“Systematic Optimization: Move beyond "vibe-based" testing. Implement rigorous evaluation frameworks (e.g., using LLM-as-a-judge, promptfoo, or DSPy) to measure and improve agent performance at scale.”
The ad says“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).”
The ad says“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).”
The ad says“Systematic Optimization: Move beyond "vibe-based" testing. Implement rigorous evaluation frameworks (e.g., using LLM-as-a-judge, promptfoo, or DSPy) to measure and improve agent performance at scale.”
The ad says“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.”
The ad says“Agent Architecture: Design and implement multi-agent workflows, including task decomposition, state management, and tool-use (RAG, API integration, etc.).”
The ad says“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).”
The ad says“Agent Architecture: Design and implement multi-agent workflows, including task decomposition, state management, and tool-use (RAG, API integration, etc.).”
The ad says“Agent Architecture: Design and implement multi-agent workflows, including task decomposition, state management, and tool-use (RAG, API integration, etc.).”

About us:

At Pencil, we are driving innovation in advertising technology through our state-of-the-art SaaS product, which harnesses Generative AI to redefine content creation. Our mission is to make AI the default in advertising without replacing creative people. To achieve this, we need to make sure that our technology isn’t just in the hands of big brands - we need to try to help small businesses and creative individuals too.

We’re building the agentic OS for marketing. We aren't just using integrating generative AI; we are building the machine that makes it professional, brand-safe, and scalable. We’re moving beyond simple text-in/text-out interfaces toward complex multi-agent architectures that can handle the nuanced demands of global brands and the agility of small businesses.

We are looking for an Agent Architect who thinks in systems, not just sentences. You will be responsible for the "brain" of Pencil—designing the agent logic, tool-calling structures, and evaluation loops that power our core creative engine and bespoke client solutions.

The Role: Architecting the Creative Brain

You won’t just be "prompting"; you will be engineering behavior. You will bridge the gap between creative intent and machine execution, ensuring that our agents are robust, predictable, and capable of high-fidelity output across text, image, and video. This is a client-facing technical role: you’ll spend as much time in rooms with brand and agency stakeholders as you will in the system. You need to be able to lead a discovery conversation with a non-technical client, decompose their problem into an agent architecture, and explain your design choices in plain language.

Your work will fall into two high-impact pillars:

Core Systems: Designing and scaling the foundational agents that power the Pencil SaaS platform.
Client Solutions: Architecting custom workflows for world-class brands that require specific "brand DNA" and complex creative logic.

Key responsibilities:

Agent Architecture: Design and implement multi-agent workflows, including task decomposition, state management, and tool-use (RAG, API integration, etc.).
Systematic Optimization: Move beyond "vibe-based" testing. Implement rigorous evaluation frameworks (e.g., using LLM-as-a-judge, promptfoo, or DSPy) to measure and improve agent performance at scale.
Scalable Frameworks: Develop reusable agents and prompt libraries that allow the platform to serve 1,000+ brands with unique voices simultaneously.
Cross-Functional Engineering: Partner with AI Engineering and Product teams to determine which behaviors should be handled via prompting, RAG, or fine-tuning.
Client Architecture: Act as the technical lead for complex client deployments. Run structured discovery sessions with clients, decompose ambiguous business problems into buildable agent workflows, and present trade-offs (cost, latency, reliability) to non-technical stakeholders in plain terms.

Your background:

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.

Benefits:

25 days PTO plus public holidays, although we operate a Flexible Time Off scheme
Health insurance / private medical cover
Monthly stipend towards wellness, fitness, and learning and development
Remote - work from anywhere in your home country
Enhanced parental leave policies, whether you become a parent through birth, adoption or surrogacy
Access to our Pencil office in The Shard, London for UK employees
Flexible working hours