Job Description

Thesis is hiring an AI Engineer to build and deploy reliable AI agents, integrations, and internal tools. The role partners with teams across the business to improve workflows, make company knowledge and data more useful, and ensure AI systems are adopted and continuously improved.

Responsibilities

Design and ship agents that retrieve information, use tools, and complete multi-step tasks across our business systems
Connect workflows across tools such as Slack, Google Workspace, Asana, Shopify, and Klaviyo through APIs, webhooks, and MCP integrations
Make long-running workflows dependable with persistent state, retries, duplicate prevention, clear approvals, and recoverable failures
Partner with Creative and Growth to improve the path from customer research and performance insights to briefs, scripts, creative variants, and testable landing pages
Build workflows that combine brand knowledge, approved product claims, asset libraries, and creative performance data while keeping human review where it matters
Help teams learn from prior experiments so each new brief or test benefits from what we've already discovered
Extend our shared skills, knowledge base, retrieval, and context-management systems so AI tools can find and use the right source material
Partner with our data team to connect AI workflows to governed analytics in Metabase and Hex, backed by Snowflake, rather than inventing competing definitions of business metrics
Use SQL and data validation to investigate issues, verify outputs, and make analytical workflows easier for non-technical teammates to use
Build evaluations using real business tasks. Test accuracy, task completion, tool behavior, failure recovery, latency, and cost before scaling a workflow
Add monitoring, useful logs, version control, automated tests, and clear documentation. Own failures after launch and fix their causes
Protect customer and company data, separate brand and account contexts, and preserve human approval for consequential actions
Choose the simplest approach that works, including ordinary code or deterministic automation when an LLM adds no value
Work directly with teammates to understand their workflows, ship usable interfaces, and help them adopt what you build
Create reusable skills and integrations rather than isolated prompts that only their author can operate
Evaluate new models and tools through practical experiments, then explain what is worth adopting and why
Measure outcomes such as time saved, adoption, output quality, and faster experiment cycles, not just the number of automations shipped