About the Role
PalUp is building an AI sales agent for merchants. Most agent work gets judged on how a demo feels — ours doesn't. There's no agent framework here waiting for you to fill in. You'll run it end to end yourself: watching how real shoppers behave, research, prompt design, implementation, and verification.
What You'll Do Every Day
Build the agent's core capabilities end to end. From understanding what merchants and shoppers actually need, through design and implementation, to verifying the result once it's live.
Ship fast with Claude Code. You'll lean heavily on Claude Code to turn complex agent logic into working product in hours.
Refine the experience from several angles. The merchant's, the shopper's, and the commercial outcomes. The same design decision often has three different right answers depending on which one you're looking through.
Make the agent measurably better over time. Iterate against what the control group actually shows, not against how it feels.
Who We're Looking For
Claude Code is how you build. Not an occasional assist — it's your default way of turning an idea into working software.
You've built or improved an AI agent in practice — a sales agent, a customer success agent, or any autonomous agent. You're fluent in one or more of: prompt engineering, RAG, agent architecture, tool use, multi-agent orchestration.
You have a track record of getting things done. It doesn't have to be code — it might be a business you got off the ground, a community you grew to real activity, or a system you designed that made something run smoothly. What matters is evidence that you find a way when there's no playbook.
You communicate clearly. Designing an agent is thinking in language: if your instructions are vague, so is the agent.
You're very curious. When something an Agent does doesn't make sense, you keep digging and testing until you understand it — because you want it to be better.
You want real ownership. You want to build this agent as your own product, not execute someone else's spec.