Invisible Technologies
Last reviewed
Jun 8, 2026
Sources
5 citations
Review status
Source-backed
Revision
v1 · 1,299 words
Improve this article
Add missing citations, update stale details, or suggest a clearer explanation.
Suggest edit
CtrlK
Last reviewed
Jun 8, 2026
Sources
5 citations
Review status
Source-backed
Revision
v1 · 1,299 words
Add missing citations, update stale details, or suggest a clearer explanation.
Invisible Technologies is an American artificial intelligence company that combines a global network of vetted human experts with its own orchestration software to provide AI training data, reinforcement learning from human feedback (RLHF), model evaluation, and enterprise process automation. Founded in 2015 by Francis Pedraza, the company began as a "digital assistant" and business process outsourcing platform before pivoting into AI training and operations, a transition catalyzed when OpenAI engaged it for RLHF work in 2022. It is frequently described as a rival to Scale AI, though it pairs its data business with an "agentic operations" arm that automates enterprise workflows. In September 2025, Invisible raised $100 million at a reported valuation of more than $2 billion.[1][2]
Invisible Technologies sells what it describes as an end-to-end AI software platform, branded Meridial, that ingests and structures messy data, maps and builds digital workflows, deploys AI agents, evaluates model and process performance, and mobilizes human domain experts where automation alone is insufficient.[1] The company says it has provided training and evaluation work for more than 80 percent of the world's leading AI model providers, naming customers such as Cohere, Microsoft, and Amazon Web Services among its clients.[1][2]
The business sits at the intersection of two markets: the AI data market served by firms like Scale AI and Surge AI, where labs pay for human-generated training and preference data, and the enterprise automation market, where companies pay Invisible to run defined, outcome-based workflows. Invisible operates a distributed model, employing roughly 350 full-time staff alongside a marketplace of more than 3,000 specialist contractors across 35 or more countries.[3] It is headquartered in New York, with additional offices in San Francisco, Washington, D.C., and London.[1]
Invisible Technologies was founded in 2015 by Francis Pedraza. Its original product was a remote "digital assistant" and outsourcing service built on what the company called a "digital assembly line": business processes were decomposed into small, repeatable micro-tasks and distributed to a network of trained human agents, with software coordinating the handoffs.[3] Early backers included Backed VC, which led a 2018 seed round, along with later investors such as Greycroft and angel investors. The company landed a breakthrough enterprise engagement with DoorDash in 2020 during the COVID-19 pandemic, and reported a roughly 24-fold increase in revenue between 2020 and 2023.[3]
The pivotal shift came in 2022, when OpenAI engaged Invisible to supply RLHF services, contributing human feedback used to fine-tune large language models in the period surrounding the launch of ChatGPT. That engagement repositioned the company from general process outsourcing toward specialized AI training and evaluation, and led to additional work with Microsoft, Amazon, and Cohere.[2][3]
Pedraza built the company with an unusual remote-first, partnership and ownership-oriented culture rather than a conventional venture-scaled startup, and Invisible operated profitably for much of its history. In January 2025, the company appointed Matthew Fitzpatrick as chief executive officer, succeeding Benjamin Plummer. Fitzpatrick previously served as the global head of QuantumBlack Labs, the AI software development and research hub at consulting firm McKinsey & Company. Pedraza moved into the role of founder, president, and executive chairman.[4]
Invisible groups its offerings around its Meridial platform and its expert marketplace. The platform is organized into several named modules:[1]
| Component | Function |
|---|---|
| Neuron | Data infrastructure for integrating and transforming structured and unstructured data |
| Atomic | Process mapping and workflow builder |
| Synapse | Annotation and evaluation layer |
| Axon | Agentic automation that orchestrates tasks and decisions across systems |
| Expert marketplace | On-demand access to vetted human specialists for RLHF and validation |
For frontier AI labs, Invisible supplies the human-generated data used to train and align models. Contributors write demonstrations, compare and rank model outputs, and rate responses for accuracy, helpfulness, and safety, the core inputs of RLHF and related preference-tuning methods. This work overlaps directly with data labeling and data annotation services offered by competitors, but Invisible emphasizes higher-skill, domain-expert annotation across fields such as engineering, finance, and healthcare.[1][3]
The company runs model evaluation and validation, including building reinforcement learning "gyms" and task simulation environments in which models are tested against realistic tasks. In March 2026, Invisible agreed to acquire WeCP, an AI-native technical assessment platform with a library of more than 18,000 scope-specific assessments and over two million real-world interview records. Invisible said the acquisition would strengthen its ability to evaluate and source experts and accelerate its RL gyms and simulated environments, integrating WeCP into the Meridial platform. WeCP was founded by Abhishek Kaushik and Mohit Goyal.[5]
Beyond data, Invisible sells "agentic operations," combining its Axon orchestration layer with human experts to automate enterprise business processes end to end. Rather than selling raw labeling hours, the company structures engagements around defined processes and outcomes, deploying AI agents for routine steps while routing exceptions to human reviewers.[1][3]
Invisible was relatively bootstrapped and profitable for much of its existence, raising comparatively little outside capital before 2025. On September 16, 2025, it announced a $100 million growth round led by Vanara Capital, the first publicly announced deal for the firm, which had recently spun off from private equity group TPG. Additional investors in the round included Princeville Capital, HOF Capital, Freestyle VC, Acrew Capital, Greycroft, Backed VC, and others. The round brought the company's total capital raised to roughly $144 million.[1][2]
Bloomberg reported the round valued Invisible at more than $2 billion, up sharply from a roughly $500 million valuation in 2024; the company itself did not formally disclose the valuation in its announcement.[2] Invisible reported approximately $134 million in revenue for 2024, more than double its prior-year figure of about $60 million, and was ranked the second fastest-growing AI company on the 2024 Inc. 5000 list. Third-party estimates put the company's 2024 EBITDA in the region of $15 million, indicating continued profitability.[2][3]
The company's commercial model is largely usage- and outcome-based: enterprise clients pay for defined processes and results, such as per-thousand-annotation rates or completed workflows, rather than purely for labor hours.[3] Proceeds from the 2025 round were earmarked for research and development on the core platform, with the engineering organization roughly doubling in size during 2025.[1]
Invisible competes in an increasingly crowded market for human-in-the-loop AI data and operations. Its closest comparisons are AI data specialists such as Scale AI, Surge AI, Mercor, Turing, and Toloka, along with established data-services vendors like Sama and Appen and large traditional outsourcers including Accenture and TaskUs.[2][3] Whereas Scale AI is most associated with data labeling and model data products, Invisible differentiates itself by coupling its data and RLHF business with ongoing, software-orchestrated enterprise operations.[2]
The company's trajectory illustrates the strategic importance of high-quality human data and expert judgment to modern AI development. As frontier labs exhaust easily available public data and shift toward post-training, alignment, and agent evaluation, demand for skilled human feedback and realistic task environments has grown, making the "human-plus-software" data and operations category a significant and contested part of the AI supply chain.[2][3] Invisible's positioning, spanning both AI-lab training data and enterprise AI enablement, reflects a broader convergence of the AI data, evaluation, and automation markets.