Labelbox Company Spotlight: Experts, Environments and Agents

Meta's GIM research thanks Labelbox for annotation support and describes its work organizing participants for a human-plus-AI study. Recruiting people, checking their qualifications and handling their participation are concrete parts of an AI infrastructure business that can be easy to miss behind the model on the screen.

Labelbox is a San Francisco company founded in 2018 by Manu Sharma, Brian Rieger and Dan Rasmuson, with Manu Sharma serving as CEO. Its current products span Horizon, environments for training and evaluating AI models, and Recursion, a platform for agents carrying out enterprise work. Human expertise supplies a connection between the company's earlier data business and those newer products.

An October 14, 2026 Series E financing package of more than $100M, combining equity and credit, supports that expansion. The company is also hiring across environment engineering, research and expert recruitment. Those choices make Labelbox useful to study as a business coordinating software, practical work and the people capable of judging its results.

The Human Expertise Inside Labelbox's Infrastructure

The GIM research paper supplies a public example of Labelbox's role. Its acknowledgments name the company for annotation support, while its study protocol describes a contract under which Labelbox recruited and vetted participants, handled compensation and collected consent. The Meta research team set the criteria and independently scored the resulting responses.

That separation makes the work legible. A researcher can specify the people a study needs and the conditions under which they should work, while another organization manages the practical arrangements required to bring them into it. The paper establishes that contribution without making Labelbox the author of the benchmark or the owner of its results.

Labelbox's own customer material describes a broader data-production role for GIM and a human-preference evaluation service for Google Cloud's Vertex AI. These are company accounts of its work. They show the kinds of relationships Labelbox seeks with research and enterprise customers, while claims about service speed and effectiveness remain the supplier's description.

In February 2026, Labelbox acquired Upcraft, whose sales-agent technology was intended to help recruit, engage and support domain experts in Alignerr. Applying sales automation to the supply of human judgment is a specific operating choice. The software business still needs people with the expertise required for the work it sells.

Horizon and Recursion Put That Experience to Different Uses

Horizon serves model developers with reinforcement-learning environments and evaluations for reasoning, tool use and computer use. WorldSim recreates enterprise applications and business data, allowing tasks to cross the systems involved in an actual workflow. The offering includes correctness checks and human preference judgments, which assess different qualities of a model's attempt.

For an AI laboratory, designing the task and its reward is part of teaching the model. A check has to capture something useful about the intended work, and the environment has to make the attempt possible. Expertise enters through the problem chosen, the conditions built around it and the way the result is assessed.

Recursion Managed Agents addresses the enterprise team that needs work completed. Labelbox describes agents starting from schedules, messages or API calls, operating through scoped tools and receiving grades against a customer's criteria. Coordinators can assign specialist agents, with session records and memory carrying information across the process. Those are the company's stated product capabilities.

The two markets create different obligations for the same supplier. A lab may want an exercise that exposes a model's limits; an operating team needs a result it can use inside its business. Labelbox's experience organizing expert judgment is relevant to both, although a successful training task cannot by itself establish the reliability of an enterprise deployment.

Founders, Capital and the Company's Stated Operating Style

The founding team came together around training data, after encountering the difficulty of turning collected information into something machine learning could use. Manu Sharma's current company profile identifies him as co-founder and CEO. Brian Rieger is a co-founder; Dan Rasmuson is a co-founder and former CTO, now listed by Humata as its founder and CTO.

Labelbox's Series E announcement names Infinitum Partners as lead investor, with Wipro Ventures, a16z, B Capital, Tau Ventures and Snowpoint Ventures participating. The stated valuation is $1.25B pre-money, and the division between equity and credit is undisclosed. The company plans additional research, engineering, customer-facing staff and compute for Horizon and Recursion.

Labelbox also reports more than 400 customers and says it reached profitability in 2026 while growing more than 5x year over year. The growth measure is unspecified, and the announcement does not disclose audited revenue or profit figures.

Labelbox describes its values as customer outcomes, seeking to understand and craftsmanship. Its careers material emphasizes individual ownership and quick decisions. These are the employer's stated expectations rather than an independent assessment of the workplace, but they offer context for the people it is looking to hire into a business spanning research, product development and customer delivery.

What Labelbox's Current Roles Say About the Work Ahead

Labelbox's live jobs board lists roles including Forward Deployed Engineer, RL Environments; Forward Deployed Research Scientist; Staff ML Engineer, Agent Training & Environments; and Global Specialized Domain Expert Recruiter, a contractor position. The engineering and research examples are in the San Francisco Bay Area, while the recruiter listing is remote in the United States. Availability can change, so the current board remains the place to inspect the roles.

The combination supplies a modest but concrete hiring signal. Labelbox is seeking people who can build environments, bring research into customer work and source specialist expertise. An opening describes a capability the company wants to add; it does not establish how many people have joined, a hiring-growth rate or a measured level of customer demand.

That operating context distinguishes the company from a simple description of an AI software platform. Its products depend on tasks, evaluations and expertise being organized well enough for another system to use them. A forward-deployed researcher and an expert recruiter can touch different parts of that process while contributing to the same customer relationship.

The careers page offers the employer's fuller description of its environment, and the active listings make its current needs more specific. As Horizon and Recursion expand, Labelbox's next team members will be joining work that crosses the simulated task, the customer's software and the human expertise needed to assess what happened there.