Description
Summary What we are building We build AI-native products where Claude is the core of the product, not a bolt-on chatbot. Think agentic workflows, internal tools that replace manual ops, and customer-facing features powered by the Claude API. We are looking for an engineer who lives in this stack daily and can go from a rough idea to a working, shipped feature fast. What you will do *Design and build Claude-powered product features using the Claude API (Messages API, tool use, streaming, prompt caching, structured outputs) *Build multi-step agents with the Claude Agent SDK and Claude Code, including custom tools, subagents, and skills *Write and connect MCP servers so Claude can read and act on our real systems (databases, CRMs, internal APIs, file stores) *Automate business workflows end to end using n8n, Make, or Zapier, with Claude in the loop for reasoning steps *Own prompt architecture: system prompts, context management, evals, guardrails, and cost and latency tuning *Ship production code (Python or TypeScript), with clean APIs, logging, error handling, and retries *Instrument everything: token spend, success rates, failure modes, and a simple eval harness so quality does not drift You are a fit if you *Have shipped at least 2 production products or internal tools built on Claude or a comparable LLM stack, and can walk us through them *Know the difference between a prompt that demos well and an agent that survives real users *Are fluent with tool use, function calling, RAG, vector databases, and when NOT to use RAG *Write solid Python or TypeScript, not just glue code *Move fast, communicate in writing, and ask sharp questions instead of guessing Experience with OpenAI or Gemini for comparison and fallback routing Vercel, Supabase, AWS, or GCP deployment experience Voice agents (Vapi, Retell, ElevenLabs) or browser automation (Playwright, Puppeteer) Experience with evals frameworks and LLM observability (LangSmith, Langfuse, Helicone)