AI原生运营主管
Head of AI Native Operations
关于 SUPABASE
Supabase 是一个基于 Postgres 的开发平台，由开发者为开发者构建。我们提供完整的后端解决方案，包括数据库、身份验证、存储、边缘函数、实时功能和向量搜索。所有服务深度集成，专为增长而设计。
关于该职位
我们正在寻找一位 AI 运营负责人，负责 Supabase 的运作方式。
Supabase 正在构建一个原生 AI 的内部操作系统：公司内统一的工作方式，其中 AI 承担相当一部分的运营负载，而不是作为助手在一旁辅助。我们将建立一个新的核心团队来构建这些系统，赋能其他团队，并将 AI 运营嵌入整个组织。这就是这个职位所要负责的。
你将负责评估团队的成熟度模型、定义工作方式的标准以及运营节奏，在一个设计原则下：每个定期会议都必须说明为什么自动化报告不能替代它。你将领导一个由高级个人贡献者组成的小团队，其中一人负责构建平台，另一人负责赋能。
其核心理念是：以代理形式编码的流程比依靠人力执行的流程更具扩展性。你的任务是证明这是一种可重复的模式，而非偶然现象，跨越你没有直接管理权的各个团队。
我们是完全远程和异步工作的。文档是这里的执行机制，这意味着标准本身就是你的交付成果。
你将负责以下内容
在这个职位中，你将：
- 全程负责成熟度模型：评分标准、评分方法、证据标准和申诉路径。每个团队都将获得一份公开的评估结果和一份通往下一阶段的书面计划。
- 开展评估和承诺：亲自与部门负责人进行评估，将它们转化为有指定负责人和截止日期的承诺，而不是意见。
- 负责标准层：定义工作如何被跟踪、决策如何被提出和解决、团队如何迁移到共享工具，以及过时文档如何被淘汰。标准以文档形式发布，团队遵循即可，无需追着执行。
- 优化决策流程：确保书面提案得到解决，而不是堆积未决，通过定义阶段、退出标准和关闭机制来实现。
- 管理运营节奏：主导自动化的每周报告、定期的障碍与决策论坛、每月的领导层审查，以及每季度的董事会级别汇报。删除那些可以被报告替代的会议。
查看英文原文
ABOUT SUPABASE
Supabase is the Postgres development platform, built by developers for developers. We provide a complete backend solution including Database, Auth, Storage, Edge Functions, Realtime, and Vector Search. All services are deeply integrated and designed for growth.
About the Role
We are looking for a Head of AI Native Operations to build the autonomous systems that run Supabase.
Supabase is building an AI-native internal operating system, where agents carry the operational load end to end. We are standing up a new central team to build that platform and run it in production. This is the role that leads it, and it is hands-on. You design the architecture, build the first versions of the critical pieces, and are accountable for whether every agent we run is correct, safe, and improving over time. We build rather than buy, and there are no settled best practices for most of this yet, so you will be setting them.
You will lead a team of senior individual contributors extending the platform and driving adoption across teams.
What You’ll Be Responsible For
In this role, you’ll:
- Own the agent platform architecture: The harness and orchestration layer and connectors into company systems, shared context and memory, and the path from prototype to production.
- Catch regressions before they ship: Model, prompt, and tool changes can’t ship until they pass the evals. When a new model comes out, you know within hours whether to adopt it.
- Make every agent observable: Every run and every decision is traced. For any agent at any time, you can answer what it did, why, what it cost, and how its quality is trending.
- Own safety and trust: Design the controls that let agents act without a human in the loop: least-privilege credentials, hard policy limits, escalation thresholds, audit logs, sandboxing, and prompt-injection defenses.
- Replace workflows, don’t assist them: Pick the workflows agents can fully own today and take them to production.
- Turn wins into a platform: Package what works into reusable components, templates, and eval harnesses so other teams can ship their own production agents without starting from scratch.
- Build and direct the team: Hire and grow senior ICs, review their designs and pull requests, and hold the quality bar.
- Measure what matters: Track workflows automated, eval pass rates, cost per run, and work absorbed against the hiring plan.
How You’ll Think
A Point of View on Where Automation Wins
You have a thesis about how AI and automation change the way a company operates, and it comes from building, not reading. You can look at a business and say which workflows agents can own today, which they will own in 12 months as models improve, and which still need people, and why. You defend that view with evidence from systems you have shipped, and you update it when the evals tell you to.
Eval-Driven
You don’t trust an agent because the demo looked good. You trust it because it passes a test set of real cases, and you know which cases it fails. When anything changes, your first question is “what do the evals say?”
Inversion Thinking
Left alone, an autonomous system tends to fail quietly, or people route around it. So the first question about any agent isn’t “does it work on the happy path?” but “how will this break, and how will we know?” The design follows from that answer: what the agent is never allowed to do, what it escalates, and which alert fires when quality drifts.
Recursive Thinking
Every system you build should produce the evidence that grades it. An agent isn’t done when it works. It’s done when it tells you, without you in the loop, that it has stopped working.
Must Have
- Hands-on agent building: Have personally designed and built autonomous agent or LLM systems that ran in production and that others depended on. Prototypes, demos, and configuring a vendor tool don’t count. You don’t need to be a career software engineer, but you must have built these systems yourself.
- Architecture depth: Can explain how the pieces fit together (orchestration, tool and data access, context and memory, tracing, evals, guardrails, deployment), with opinions on current frameworks.
- Security judgment: Have shipped agents that take real actions with real permissions, and designed the controls that made that safe.
- A tested point of view: Can argue, with examples from your own builds, where autonomous systems create real business leverage and where they don’t yet.
- Technical leadership: Have led senior builders and set the quality bar through design and code review.
Strong Signal
- Have replaced a human-run business workflow (finance, operations, support, release management) end to end with an autonomous system.
- Are native to Git, Markdown, and pull requests, and comfortable reading and writing Python or TypeScript.
- Have taken an internal platform from zero to adoption by teams you did not control.
- A software engineering background is a plus, but not required.
What Success Looks Like
- Six months: 20+ autonomous agents in production on the shared platform, each with an eval suite, tracing, and a named owner.
- Regressions get caught, not discovered: Quality problems surface in an alert or dashboard before a user reports them.
- Workflows replaced, not assisted: Multiple high-volume workflows are fully owned by agents, with humans handling only escalations.
- Trusted by default: No agent causes material financial, security, or customer harm, and every agent’s permissions and actions can be audited.
- Teams ship their own: At least three teams outside the core team run production agents built on the platform.
- Autonomous team: Your two reports work independently, and you spend your time on architecture and review rather than execution.
WHAT WE OFFER
- Fully Remote
We hire globally. We believe you can do your best work from anywhere. There are no Supabase offices, but we provide a WeWork membership or co-working allowance you can use anywhere in the world.
- ESOP
Every team member receives ESOP (equity ownership) in the company. We want everyone to share in the upside of what we’re building together.
- Tech Allowance
Use this budget to set up your ideal work environment—laptop, monitor, headphones, or whatever helps you do your best work.
- Health Benefits
Supabase covers 100% of health insurance for employees and 80% for dependents, wherever you are. Your wellbeing and your family’s health are important to us.
- Annual Off-Sites
Once a year, the entire company gathers in a new city for a week of connection, collaboration, and fun. It’s a highlight of our year.
- Flexible Work
We operate asynchronously and trust you to manage your own time. You know what needs to be done and when.
- Professional Development
Every team member receives an annual education allowance to spend on learning—courses, books, conferences, or anything that supports your growth.
ABOUT THE TEAM
Supabase was born-remote and open-source-first. We believe our globally distributed team is our secret weapon in building tools developers love.
- ~400 team members
- 60+ countries
- 20+ languages spoken
- Over $1B raised (including our $500M Series F)
- 540,000+ community members
We move fast, build in public, and use what we ship. If it’s in your project, we probably use it in ours too. We believe deeply in the open-source ecosystem and strive to support—not replace—existing tools and communities.