> Markdown version of [/jobs/ext/3970954-ai-engineer](https://www.wearedevelopers.com/jobs/ext/3970954-ai-engineer). Every page supports `.md` or `Accept: text/markdown`. Links point to the HTML versions so they work for humans too. Agent guide: [/agents.md](https://www.wearedevelopers.com/agents.md). --- # AI Engineer - **Company:** The Arcade - **Location:** San Francisco, CA, United States - **Experience:** Experienced - **Salary:** $230,000.0 - **Contract:** Permanent contract - **Skills:** Application Programming Interfaces (APIs), Artificial Intelligence, User Authentication, Data Warehousing, Programming Tools, Distributed Systems, Python (Programming Language), Microsoft Software, Software Engineering, Systems Integration, TypeScript, Design of Telemetry/logging Agents, Large Language Models, Production Code, Free and Open-Source Software, Model Context Protocol, Pagination, Splunk, GPT, Api Management - **Published:** October 27, 2026 - **Apply:** https://www.careerbuilder.com/job-details/ai-engineer-san-francisco-ca--739b1780-a849-4b81-be61-61d569e9cebc ## About the Role Artificial Intelligence (AI), Customer/Client Research, Product Demonstration, Artificial Intelligence (AI), MCP - Microsoft Certified Professional, Authentication, Application Programming Interface (API), Splunk, Distributed Computing, Microsoft Product Family, * 4+ years of software engineering experience shipping production code, with a track record of owning projects end to end. * Strong Python and/or TypeScript * Experience building and consuming APIs at scale - REST, auth flows, pagination, rate limits, and the messy parts of real-world integrations. * A real desire to build the parts of this stack that don't exist yet - comfortable when the answer isn't known and you have to figure it out. * LLM application experience - prompting, retrieval, tool use, or agent design. * A testing mindset - tests and evals that catch real failures, not just green checkmarks. * Comfort with ambiguity - early team, a narrow charter that will expand, decisions made with incomplete data. * Stop-at-nothing energy to get the job done. Bonus Points * Familiarity with the MCP (Model Context Protocol) ecosystem or similar agent-tool protocols - extra bonus if you've filed an issue against the spec. * Familiarity with Go * You've built evals or measurement systems for ML/AI behavior. * Prior experience at an API platform, data-warehouse, integrations-heavy product, or developer tools company. * Open-source contributions. * Experience at an early-stage startup, and you loved it. ## Description You'll report to the Engineering Manager for Abilities, Builder, and Growth. Abilities exists to enable agents to get things done - full stop. We take a job to be done that someone's trying to do (or that we want to enable) and figure out how to deliver it, whatever that takes: updating the Arcade Engine itself (Go), building an MCP server, standing up an entirely new service to support Skills, building an agent and figuring out how to ship it via MCP, or whatever's next in this space. You're building the tools, skills, and platform underneath them - the pieces that let an agent do things nobody's figured out how to do yet. Every time an agent hits a wall in Cowork or ChatGPT - some action it just can't take - that's a rough edge the Abilities Team gets to fix, at the scale of every customer who hits it. The team is early and the surface area is growing fast, so you'll have real ownership from day one, including a hand in the patterns every toolkit author after you inherits, and in agents-as-tools (sub-agents), which Abilities also owns. What You'll Do * Build the systems that build systems - dedicated agent harnesses that write and maintain tools themselves, plus agents-as-tools (sub-agents) that let one agent call another as a capability * Take a job to be done and own delivering it - whether that's changes to the Arcade Engine, a new MCP, a new service, or something nobody's built yet. * Build new toolkits and abilities - turn a vendor's API (or an internal need) into tools, skills, or agent abilities that get customers a real outcome, not just a demo. * Fix the rough edges - when an agent hits a wall, figure out what's missing and build it, at the scale of every customer who'll hit that same wall. * Maintain and improve the existing catalog - fix breakages, sharpen tool descriptions and response shapes, keep quality high as upstream APIs change. * Build the next generation of evaluation and testing for abilities - go beyond pass/fail checks to know, programmatically, what a tool is actually doing across agents, models, and surfaces. * Engage with and push the MCP ecosystem, where the standards for agent tools are still taking shape.