Forward Deployed Engineer - RL Environments

Labelbox raises $100M+ Series E for Horizon and AI agents. The grading rubric is becoming part of the product a company buys from Labelbox. A manager specifying a satisfactory result gives its enterprise agents something concrete to work toward, and a way to carry corrections into the next assignment. The San Francisco company, led by co-founder and CEO Manu Sharma, announced more than $100M in Series E financing on October 14, 2026. Infinitum Partners leads the package, which combines equity and a credit facility at a $1.25B pre-money valuation. Labelbox intends to expand research, engineering, customer-facing teams and compute for Horizon, its AI training environments, and Recursion, its enterprise-agent platform. Those products bring Labelbox into two connected purchasing conversations. A frontier laboratory needs tasks that help a model improve; an operating company needs the model to finish work to its own standard. Labelbox is trying to serve both, with human judgment present in the scoring of an exercise and the evaluation of a completed job. What Labelbox's Series E financing includes. The financing announcement names Wipro Ventures, a16z, B Capital, Tau Ventures and Snowpoint Ventures alongside lead investor Infinitum Partners. Labelbox has not disclosed the allocation between equity and credit, the credit provider or how much of the facility has been drawn. The headline amount therefore describes a financing package, with different forms of capital inside it. The $1.25B figure is the company's valuation before the new equity investment. Computing a post-money valuation would require the equity component, which has not been supplied. The distinction is especially useful when comparing this announcement with rounds in which the entire stated amount purchases shares. Labelbox says it became profitable in 2026 while growing more than 5x year over year. The announcement leaves the growth measure unspecified and does not supply audited revenue or profit figures. Its plans for the proceeds center on product development, people and compute, with no disclosed budget for each category. That combination gives the financing a practical operating frame. Labelbox is supporting work for AI laboratories while expanding a product sold to enterprise teams, whose purchasing decisions depend on the work they need completed. The capital can support that expansion; the announcement provides no basis for estimating how quickly those customers will adopt it. From training data to work that can be scored. Manu Sharma, Brian Rieger and Dan Rasmuson founded Labelbox in 2018. The company initially supplied tools for capturing human expertise in training data, giving AI teams a way to organize the judgments their models needed to learn. Rasmuson is a former Labelbox CTO; Humata's current leadership page identifies him as its founder and CTO. The funding history traces the earlier business. Labelbox's company-issued Series C announcement disclosed $40M led by B Capital in February 2021. Its January 2022 Series D added $110M led by SoftBank Vision Fund II. Those are historical equity rounds, separate from today's mixed financing package. The latest product expansion changes what that expertise is used to judge. Horizon supplies reinforcement-learning environments for post-training models on reasoning, tool use and computer use. Its WorldSim system recreates enterprise software and business data so an agent can attempt a task across the systems that task involves. Correctness checks and human preference judgments provide different signals about the resulting work. An answer can satisfy a mechanical check while still disappointing the person who needs to use it. That gap gives domain expertise a continuing job as models gain more capabilities. For Labelbox, extending the training environment into enterprise software creates another place to apply the experience it accumulated organizing human feedback. What enterprise teams are being asked to define. Labelbox opened Recursion Managed Agents on October 8, 2026, following internal use. Teams specify a goal, a trigger and criteria for an acceptable outcome. A coordinator assigns work to specialists, and a separate grader examines the evidence against those criteria. Labelbox says the system can revise failed work and retain lessons for subsequent runs. The current product page describes persistent sessions, scoped tools and a record of the work performed. Those are vendor-described capabilities. The record is commercially useful because it gives the customer a place to examine how an agent arrived at its result, rather than making the delivered file the only thing available to review. Writing the rubric also exposes decisions a team might otherwise leave implicit. A finance manager may care about the explanation of a changed forecast as well as its delivery date. An engineering manager may need a reproducible defect and a useful fix, alongside a report that looks complete. These are examples of the judgment required to specify work, not claims about named customers or measured deployments. As assignments recur, the standard itself can change. A team learns what to ask for after seeing the first results, and the supplier has to make those corrections usable in the next session. The product relationship consequently includes the customer's evolving definition of an acceptable job, along with the machinery that executes it. The customer relationship behind the expansion. Labelbox's current company profile presents Horizon and Recursion as the two sides of its business. Laboratories buy environments for improving model capability; enterprises buy a system intended to apply that capability to recurring work. Both rely on a useful account of what happened and how the result was assessed, even though their immediate buyers and objectives differ. The distinction also places Labelbox in a wider enterprise-agent market. DevCuration's coverage of Scuba's shared context for teams and agents examines how organizational knowledge travels into software. Harbor's agent-risk and insurance approach brings evidence of agent behavior into a different commercial conversation. These companies address adjacent needs; the connection is an editorial observation, with no partnership implied. Labelbox's expansion adds outcome grading to that conversation. Its ability to build simulated work for laboratories supplies relevant technical experience, while enterprise adoption asks the company to understand standards that vary by team and assignment. The customer contributes a definition that has to remain useful after the model, the tools or the underlying business changes. The financing gives Manu Sharma's team more resources to develop both products. An enterprise customer's next correction is still consequential: it can change the rubric, the evidence the grader asks for and the memory an agent takes into its next job. Labelbox's relationship with that customer continues through those revisions, one assignment at a time. DevCuration Data AI Infrastructure funding, last 30 days. DevCuration's funding database tracked 37 AI Infrastructure rounds totaling $13.9B in disclosed capital over the past 30 days. Recent deals DevCuration covered:

Labelbox raises $100M+ Series E at $1.25B valuation to advance RL and AI Agents - unite.ai. Labelbox said on October 14, 2026, that it raised more than $100 million in Series E financing at a $1.25 billion pre-money valuation, in a round led by Infinitum Partners that combines equity with a credit facility. Wipro Ventures, Andreessen Horowitz (a16z), B Capital, Tau Ventures, and Snowpoint Ventures also participated, among others. Chief executive and co-founder Manu Sharma detailed the terms in a company blog post, and the company said the financing will support continued research and engineering across its two core products, Horizon and Recursion Managed Agents, as well as expansion of the teams bringing those products to customers. Sharma wrote that access to compute will be central to the company's next chapter, noting that Infinitum's John Yetimoglu was an early investor in Groq and is now involved in building large-scale compute infrastructure at GroqCloud. Company position and history. Labelbox stated that it reached profitability in 2026 while growing more than fivefold year over year. The company reports more than 400 customers, including over 90% of leading U.S. AI labs, and says its base spans frontier AI labs, hyperscalers, startups, and Fortune 500 companies. Sharma, Brian, and Dan founded Labelbox in 2018. In a founder's note, Sharma wrote that he came to the United States to learn to design and build aircraft and rockets, working on computational fluid dynamics and neural networks, and that his later roles at DroneDeploy and Planet Labs showed him how much of the opportunity in machine learning sat in data. The post describes the company's history in three chapters: training-data software for people building AI; reinforcement-learning environments built after what Sharma calls the ChatGPT moment, when the most capable models began improving by practicing real work; and agents, which the company built for itself first before opening the system to customers. Sharma also wrote that Labelbox builds each layer of its platform itself: an agent harness designed for ultra-long-horizon agents, sandbox infrastructure that runs millions of virtual machines, its own reinforcement-learning post-training infrastructure, and an expert network of more than 4 million people. Horizon reinforcement-learning environments. According to the company, Horizon's WorldSim system builds simulated companies, with code, dashboards, spreadsheets, decks, and documents filled with realistic business data, in which a model attempts a difficult multi-step task end to end while every step is recorded as a trajectory. Deterministic verifiers score each trajectory the same way on every run, and when two trajectories are both correct, an expert's preference teaches the model what Sharma calls taste. The company describes operating at a scale of thousands of worlds and millions of graded trajectories, tuned to a lab's pass@k targets across coding, computer use, research, multimodal, and voice work. The Horizon product page lists domains that include autonomous AI research, agent coding and software engineering, multimodal knowledge work, voice with agentic tool use, computer use, and cybersecurity. Sharma wrote that Labelbox now works with 90% of the leading frontier AI labs and hyperscalers. Recursion Managed Agents. The funding announcement states that Labelbox opened Recursion Managed Agents to enterprise customers earlier in October 2026. Under the product's design, a Managed Agent receives a goal, a trigger such as a schedule, an alert, a Slack message, a webhook, or an API call, and a rubric defining what good work looks like. A separate grader agent scores every run against the rubric line by line and points to evidence for each call; when the work misses, the agent revises it. Agents can operate in three modes: multi-agent, in which a coordinator chains specialists that each carry their own model, instructions, tools, and permissions; team, in which copies of an agent split work on a shared board and check each other's results; and auto, the default, in which the agent decides for itself when a job needs specialists or a team. The company says agents work across more than 100 business applications and any MCP server, with every tool scoped as read only, changes data, or destructive and credentials held in a vault so the model never sees them. Sessions run in isolated sandboxes, and Labelbox says the platform runs with zero data retention at every LLM provider. Its listed compliance program covers SOC 2 Type II, GDPR, CCPA, NIST 800-171, and PCI DSS. Pricing is set per agent session hour, with compute and model spend combined in one estimate, and existing Labelbox enterprise customers receive a one-time $100 credit. Describing the company's internal rollout in a product launch post dated October 8, 2026, Labelbox wrote, "The surprise wasn't that agents could do the work. It was how quickly they got better at it." The company reports running 51 agents across 11 teams, spanning software engineering, reliability, security, machine-learning research, evals, product, finance, sales, strategy, and procurement. In one internal example, Labelbox said its pull-request review agent ran 39 graded reviews on a single agent version and model, and that across the 33 reviews that passed the grader, the median cost fell from $13.15 for the first 11 reviews to $8.90 for the last 11, 32% less per review, while the agent's memory was updated 15 times and time per review held at about 27 minutes. Under the product's memory mechanics, once five new session reports exist, and at most once an hour, a curator agent reads them and updates the agent's memory; customers can browse what an agent has learned and delete any lesson that no longer holds.

Micro1 hits $500M run rate as AI training data gold rush accelerates. AI data startup Micro1 reaches $500M gross run rate, signaling explosive growth in training data market PUBLISHED: Fri, Aug 21, 2026, 12:49 AM UTC | UPDATED: Wed, Sep 2, 2026, 1:32 AM UTC The race to feed AI models just hit a new milestone. Micro1, a startup specializing in AI training data and data labeling services, has reached $500 million in gross run rate, according to TechCrunch. The achievement underscores how the AI boom isn't just enriching chip makers and model developers - it's creating massive opportunities for companies that provide the fuel these systems need to learn. As OpenAI, Google, and others race to build more capable models, the infrastructure powering that development is becoming a billion-dollar business in its own right. Micro1 just became the latest proof that in the AI gold rush, selling picks and shovels can be just as lucrative as mining for gold. The startup's climb to $500 million in gross run rate represents one of the fastest growth trajectories in the AI infrastructure space, fueled almost entirely by insatiable demand for training data. The numbers tell a story about where AI development bottlenecks really exist. While Nvidia grabs headlines with GPU shortages and OpenAI dominates conversations about model capabilities, companies like Micro1 are quietly solving a less glamorous but equally critical problem: how do you generate enough high-quality, human-labeled data to actually train these models? The answer involves armies of human annotators, sophisticated quality control systems, and increasingly complex workflows for reinforcement learning from human feedback - the technique that helped make ChatGPT feel conversational. Micro1's platform connects AI companies with skilled workers who label images, rate AI responses, and provide the feedback loops that turn raw compute into useful intelligence. What makes Micro1's growth particularly striking is the timing. The company is hitting this milestone just as the industry confronts a looming shortage of quality training data. According to research from Epoch AI, we could exhaust high-quality text data for training by 2026, forcing companies to get creative about synthetic data, better labeling, and more efficient use of human feedback. That scarcity is driving up the value of platforms that can reliably deliver quality annotations at scale. The competitive landscape reflects these high stakes. Scale AI, one of Micro1's chief rivals, hit a $7.3 billion valuation in 2021 and counts the Department of Defense among its clients. Companies like Labelbox, Sama, and Appen are all vying for contracts with the big AI labs. But Micro1's approach appears to be resonating - particularly its focus on reinforcement learning workflows that have become essential for training models like GPT-4 and Claude. The business model itself is revealing. Unlike pure software companies, data labeling operations blend technology platforms with managed workforces, creating gross run rates that look impressive but come with corresponding costs. Still, reaching $500 million signals that Micro1 has found a way to scale both sides of that equation effectively. For Google, Meta, Microsoft, and other companies racing to deploy AI across their product lines, reliable access to training data infrastructure has become non-negotiable. Internal teams can only scale so far. Outsourcing to specialists like Micro1 lets them move faster while maintaining the quality controls necessary for production AI systems. The growth also highlights how AI development is becoming more industrialized. Early models could be trained by small academic teams with modest budgets. Today's frontier models require coordinating thousands of labelers, managing complex feedback loops, and maintaining consistency across millions of data points. That operational complexity creates moats for companies that can execute it well. What's particularly interesting is how this market might evolve as AI capabilities advance. Some believe that as models get better, they'll need less human feedback - potentially reducing demand for labeling services. Others argue the opposite: that as AI tackles more complex tasks, the need for nuanced human judgment in training only increases. Micro1's bet is clearly on the latter scenario. The startup's trajectory also raises questions about market saturation. At $500 million in run rate, Micro1 is already processing enormous volumes of data. How much bigger can this market get? The answer may depend on how quickly AI adoption spreads beyond tech giants into healthcare, finance, manufacturing, and other industries that will need custom training data for domain-specific models. For now, the AI training data sector looks more like early innings than late. OpenAI is preparing even larger models, Google is embedding AI across its product suite, and enterprise adoption is just beginning. Each of those trends creates demand for more data labeling, more human feedback, and more infrastructure to manage it all. Micro1 appears to be capitalizing on that wave at exactly the right moment. Micro1's climb to $500 million in gross run rate isn't just a startup success story - it's a signal about where the AI industry's real infrastructure challenges lie. As models get more sophisticated and AI deployment accelerates across industries, the unglamorous work of labeling data and managing human feedback loops has become mission-critical. The companies that can deliver that infrastructure reliably and at scale are positioning themselves at the center of the AI economy. With training data scarcity looming and model complexity increasing, Micro1's growth trajectory suggests this market has plenty of room to run. The question now is whether the startup can maintain its momentum as competition intensifies and the technological landscape shifts beneath everyone's feet. More Topics: