The posting
About HumanSignal
Real-world data is the competitive edge in AI.
HumanSignal is a human data partner for companies building AI models and products. Our customers ship better AI, faster, because we partner with their researchers from real-world data creation to annotation to delivery.
We design and create datasets from scratch, recruit and manage the domain experts who evaluate model output, and run everything through our own platform, Label Studio, the open-source standard for data labeling and evaluation, used by over 1 million practitioners worldwide.
We specialize in the operationally complex: real-world data collection, multimodal pipelines, and multi-step workflows. Advanced ML and AI teams use our enterprise platform to run their own data factories, and our services team to extend their reach where in-house capacity runs out.
If you want to do work that materially shapes how the next generation of AI products gets built, we'd love to talk.
The engineers on our team are great. We want to make sure they have a manager who makes them better, and that’s the role.
We’re building across an open-source platform, enterprise software, and a managed data-services business that runs on our own product. We need to make delivery more predictable while giving engineers more ownership – not add another layer of approvals. You’ll help shape what we build, the commitments we make, and how the team works.
You manage the team directly: 1:1s, feedback, performance, promotions, and the hard conversations. You own delivery so that what we decide to build turns into shipped software on a rhythm we can count on. You notice when someone is blocked and help them get unstuck without making yourself the dependency for every decision. When someone is struggling, you address it early.
The part of this role we care about most is helping people become better professionals: better at scoping, estimating, making decisions, and owning outcomes rather than tasks. If your instinct when an engineer struggles is to take the work back, this isn’t the role. If your instinct is to figure out what they’re missing and help them develop it, keep reading.
You still write and ship code. Not because we need another full-time individual contributor, but because I don’t believe you can effectively manage this team if you’ve stopped building. You stay hands-on to make better technical decisions, work through difficult problems, and improve how the team builds. Your job is not to become the team’s busiest engineer or the person every release depends on.
You use coding agents in real work and understand where they help, where they fail, and how to verify their output. You make the team better by showing what works and helping people apply it – not by mandating tools or mistaking more generated code for better engineering. You can challenge architectural decisions, explain the tradeoffs, and help the team choose an approach it can ship and maintain.
What you’ll own
- Delivery: Planning, scoping, and shipping quality software against a roadmap you help shape. You make tradeoffs explicit and push back when we’re asking the team to do more than it can do well.
- How the team works: Enough structure to make delivery predictable, not so much that it slows a small team down. You remove friction, address recurring blockers, and improve how we use AI tooling.
- People: Hiring, clear expectations, regular feedback, growth, promotions, and exits when they’re necessary.
- Technical judgment: Staying close enough to the code and architecture to challenge decisions and support engineers without becoming the approval gate for everything.
- The engineering response to escalations: Getting the right people involved, making priority calls, and addressing recurring causes across customer, services-team, and open-source issues. You’re accountable for the response, not expected to personally fix every problem.
What we’re building
- A ground-up rebuild of the annotation editor, with interfaces generated by coding agents instead of hand-configured.
- Native evaluation of agent trajectories: trace ingestion, replay, and hybrid human-plus-model-judge workflows.
- Continuous calibration: treating automated evaluators as instruments that can drift and need ongoing checks against human ground truth.
- The workforce platform behind our managed services business.
- Quality analytics across annotators, models, and programs.
Our own services team uses the platform to deliver customer work. You’ll see directly where the software helps, where it falls short, and which problems keep coming back. Part of the opportunity is turning that experience into better products, not just a growing list of one-off customer fixes.
Stack: Python/Django, TypeScript/React, Postgres, Kubernetes. Coding agents are part of how we work every day.
What success looks like
Engineers take greater ownership and need less intervention. Scope and delivery risks surface early, rather than at the deadline. Customer commitments and product priorities are reconciled before they become emergencies. Recurring escalations lead to lasting fixes.
The team ships more reliably, people are getting better at their work, and neither depends on you repeatedly stepping in to rescue a release.
Who we’re looking for
- You’ve managed engineers for at least a couple of years. You’ve hired and developed people, addressed underperformance directly, and made difficult people decisions when necessary.
- Before that, you were a strong engineer for long enough to develop sound technical judgment. You can reason about architecture and tradeoffs beyond your own code.
- You still build software and enjoy it. You use coding agents seriously and can explain, with examples, what effective use looks like.
- You’ve shipped at a startup and know how to make decisions with incomplete information.
- You say what you think early, give specific feedback, and don’t manage by avoidance.
- You can balance long-term product work with customer deadlines, negotiate realistic commitments, and recognize when a services request should become a product capability, and when it shouldn’t.
Nice to have: experience with ML tooling, data platforms, or evaluation systems; experience managing a distributed team; frontend architecture depth.
How to apply
Skip the form. Email me directly at [email protected] with examples of your best work. That can be code you’ve written, a system you designed, a team you built, a document that changed a decision, or a product you shipped. I want to see what you’re proud of and hear why.
You don’t need a public portfolio, and please don’t share proprietary material. A description of the problem, your contribution, and the outcome is enough. A short note on what draws you to HumanSignal specifically helps too.
I read everything that comes in. If it’s a fit, I’ll reach out.
About HumanSignal
We build the infrastructure that turns expert human judgment into reliable signal for AI systems. Label Studio, our open-source platform, is used by 250,000+ people across 60,000+ organizations. Label Studio Enterprise is the commercial version: the system of record for annotation and evaluation quality at companies shipping production AI.
Alongside the software, we run a managed data-services business that puts our own platform and expert workforce to work on customers’ hardest evaluation and labeling programs.
