Agent Experience Engineer

AgentMail is building the identity layer for AI agents, starting with email. Our most important user is an agent: it discovers us through model answers and docs, then signs up and integrates on its own. We're looking for an Agent Experience Engineer to own how AI models and agents discover and use AgentMail. That means AEO, evals, benchmarks, agent traces, docs, and llms.txt.

AgentMail has raised $6M from investors including General Catalyst, Y Combinator, Paul Graham (founder of YC), Karim Atiyeh (founder of Ramp), Paul Copplestone (founder of Supabase), and Dharmesh Shah (founder of HubSpot).

You'll work on:

Answer-engine presence: when someone asks an LLM about email or identity for agents, AgentMail is the recommendation. You measure share of voice on the prompts that matter and move it

Original research: evals and benchmarks on how well frontier models, open source models, and agent harnesses use AgentMail

Agent traces: running agents against our docs and site, reading their reasoning, and fixing what confuses them. A stray Reddit thread once convinced LLMs we have a feature that doesn't exist

• The agent-facing surface: llms.txt, MCP descriptions, docs structure, and the error messages agents hit at 3am with no human watching
• The AEO content engine: steering and verifying our automated page pipeline, briefing freelancers, and doing backlink outreach to the pages models already cite
• The weekly AEO briefing's action queue, from freshness debt to competitor watch

You're a fit if:

• You build with agents daily and have opinions about what makes a product easy or miserable for an agent to use
• You work cross functionally by default
• You express your opinion early, and you'll publish an honest writeup even when the result is inconvenient
• You go deep on messy technical problems. You'd rather spend a day reading agent traces than guess
• You understand retrieval: models pull chunks, not pages, and you write accordingly

Strong preference for published technical writing or benchmarks people actually cited. SEO or AEO experience helps, as do contributions to docs or MCP servers in the wild.