AI Product Engineer
工作时间安排：星期一~周五，早8点~晚5点 山地标准时间工作安排：完全远程工资：2000-2500美元雇主：AI 公司位于亚利桑那州尼克斯的角色 这是一项新的工程职位。你不会成为一个在后载中连接的代码猿。你将使用 Claude Code 和其他编码代理权拥有端到端功能：设计，实施，测试，运输。你指导 AI,读出它生产的每条线，然后发送完成的事情。你将跨平台工作：Threo （客户的工作空间）,Wingtip （从简单英语请求执行代理）,Pylon （代理创建和运行时机）,Canopy （每个客户租金控制飞机）. 我们的堆范围包括 .NET/C#和 FastAPI (Python) 在后端，React 在前端，以及 PostgreSQL 数据。你不需要知道一切第一天 你需要快速使用编码代理并将其运送到之前未接触过的堆中。我们聘请以品味和速度为目的。你会做什么 · 从粗略的想法来说，从设计，构建，测试，部署问题：从规格和复制级代理指令工作，你和你的代理都使用。没有任何"真正的实现". ·阅读人工智能生产的每一行。你负责什么船舶，而不是模型。· 同时运行两个，三个或更多的 AI 代理 （新功能，bug fixes, refactors）,保持每个在自己的沙箱或工作树和从他们所有同一天没有失去状态或范围。· 在 .NET 10 LTS / C# （ASP.NET 核心，最小 API,背景工人）中运输服务。· 用 FastAPI 构建和扩展 Python 服务我们的代理执行引擎和邻近工具。· 在 PostgreSQL 设计和发展数据模型：方案，迁移，索引，查询性能。· 与稳定的 React Compiler console,数据网格，表格和聊天 UI 建立 React 19 接口。· 通过工具调用和 MCP 服务器，结构化输出，反试和倒退实现流媒体聊天和实时更新，并与 LLM 提供商集成。· 写测试，编写文件，关闭循环，并在服务和堆中进行调整。AI 帮助，但它不会让你理解系统。条是"完成了",而不是"快乐的道路运行". · 通过 Docker 和 GitHub 行动将我们的云基础设施运送到我们。我们正在寻找的是 AI 流动性，判断力和速度，而非多年的经验。如果你住在克劳德代码 （或 Cursor,或 Codex）中，然后运送真实的软件。
查看英文原文
Working Schedule: Monday – Friday, 8am – 5pm Mountain Standard Time
Work Arrangement: Fully Remote
Salary: 2,000 - 2,500 USD
Employer: AI company located in Phoenix, Arizona
The Role
This is a new kind of engineering role. You won't be a code monkey wired into a backlog. You'll own features end-to-end using Claude Code and other coding agents: design, implementation, testing, shipping. You direct the AI, you read every line it produces, and you ship the finished thing.
You'll work across the platform: Threo (the client workspace), Wingtip (agentic execution from plain English requests), Pylon (agent creation and runtime), and Canopy (the control plane for every client tenant).
Our stack spans .NET/C# and FastAPI (Python) on the backend, React on the frontend, and PostgreSQL for data. You don't need to know it all on day one — you need to ramp fast with coding agents and ship in stacks you haven't touched before. We hire for taste and velocity.
What You'll Do
· Take a problem from rough idea to shipped feature: design, build, test, deploy, working from specs and repo-level agent instructions you and your agents both use. No handoffs for "the real implementation."
· Read every line the AI produces. You're accountable for what ships, not the model.
· Operate two, three, or more AI agents at once across separate work streams (new features, bug fixes, refactors), keeping each in its own sandbox or work tree and shipping from all of them the same day without losing state or scope.
· Ship services in .NET 10 LTS / C# (ASP.NET Core, Minimal APIs, background workers).
· Build and extend Python services using FastAPI for our agent execution engine and adjacent tooling.
· Design and evolve data models in PostgreSQL: schema, migrations, indexes, query performance.
· Build React 19 interfaces with the stable React Compiler — admin consoles, data grids, forms, and chat UIs.
· Implement streaming chat and live updates, and integrate with LLM providers via tool calling and MCP servers, structured outputs, retries, and fallbacks.
· Write tests, write docs, close the loop, and debug across services and stacks — the AI helps, but it doesn't excuse you from understanding the system. The bar is "that's done," not "the happy path works."
· Ship via Docker and GitHub Actions to our cloud infrastructure.
What We're Looking For
We hire on AI fluency, judgment, and velocity, not years of experience. If you live in Claude Code (or Cursor, or Codex) and ship real software with it, we want to talk.
Must Have
· Daily, fluent use of agentic coding tools (Claude Code, Cursor, Codex, or similar). We care that you can direct an agent through a non-trivial feature without losing the plot — not which tool you use — and that your first instinct is to design with AI in the loop.
· Parallel agent management. You can run multiple agents at once across separate work streams (features, bug fixes, refactors) and deliver shipped work from each.
· Solid programming fundamentals. You can read and reason about code in at least one of: C#, TypeScript/JavaScript, or Python. You understand HTTP APIs, relational databases, and basic async patterns. You can debug what you didn't write.
· Product sense. You can look at a half-formed requirement, ask the right three questions, then ship something the user actually wants.
· Ownership and clear async communication. When you pick up a feature, you finish it: tests, docs, no dangling threads, no partial solves left for someone else.
· Quality checks on AI output. You ship the full loop — traces, trajectory-level evals, and guardrails against prompt injection — and can show a non-deterministic feature is safe in front of paying customers.
Nice to Have
· Streaming systems (WebSockets, SignalR, or similar).
· LLM integration work: tool calling and MCP servers, streaming, and retrieval or context engineering.
· LLM cost and latency optimization: caching, model routing, token budgets.
· Multi-tenant SaaS architecture.
· Familiarity with MSP/PSA ecosystems (ConnectWise, Autotask, Datto RMM, etc.).
Originally posted on Himalayas