Pricing is based on agent-session hours rather than seats or a fixed subscription.
Teams should estimate workload and inspect the displayed compute and model spend before launching sessions.
Evidence source ↗
Labelbox develops Horizon reinforcement-learning environments for AI labs and Recursion managed agents for enterprise workflows, with tasks graded against user-defined criteria.
Current product capabilities and usage-based pricing are described on Labelbox's site, but no independent reviews of the newly launched Recursion Managed Agents were found. G2 feedback concerns the previous data-labeling product, so scores are cautious editorial assessments rather than user ratings or hands-on test results.
This profile combines documented product information and independent evidence where available. It is not a hands-on test unless TechZella explicitly identifies one.
Labelbox began in 2018 as software for creating AI training data. The company now presents two broader product areas: Horizon, which provides reinforcement-learning environments and evaluations for frontier models, and Recursion, which offers managed agents for enterprise work.
Recursion agents can start from schedules, events, Slack messages, webhooks, or API calls. They work across connected applications, can coordinate specialist agents, and return work graded against criteria defined by the customer. Labelbox’s public materials describe managed infrastructure, but do not list a fixed subscription price.
A quick way to understand who this product may suit, where its limits matter, and which facts are most relevant before you compare alternatives.
Enterprise teams automating recurring, multi-step work across business applications, especially when they need scheduled or event-triggered agents, rubric-based review, and a record of actions and results.
Buyers seeking a publicly priced, seat-based subscription or an independently reviewed agent product may need more information before committing; Labelbox prices agent sessions by the hour and its new offering has little public third-party review coverage, and should evaluate alternatives such as UiPath or Microsoft Power Automate instead.
API, Cloud, GitHub, Web
7.2/10
These are the product-specific differences that may matter when deciding whether the software fits your workflow. They are based on documented evidence rather than a universal ranking.
These are documented constraints or practical friction points that may matter before you adopt the product.
Teams should estimate workload and inspect the displayed compute and model spend before launching sessions.
Evidence source ↗A pilot can use the credit, but continued usage is billed under the session-hour model.
Evidence source ↗Teams need to configure credentials, access scopes, allowlists, and approval rules before automating consequential actions.
Evidence source ↗Labelbox says the pre-launch estimate includes compute and model spend. No fixed public rate is stated.
Check current pricing →Limited-time introductory credit for new teams, as stated in the Recursion launch announcement.
Check current pricing →Pricing may change. Verify current plans on the vendor's website.
Compare similar products directly, then see which alternatives may make more sense when your requirements differ.
| Decision factor | Labelbox | UiPath | Microsoft Power Automate |
|---|---|---|---|
| Pricing model | Recursion charges by agent-session hour and displays compute and model spend in an estimate before launch. | UiPath lists Basic at $25 per month and says Standard pricing is available from sales. Source ↗ | Microsoft documents free and paid user-license options, with a separate trial license for Power Automate. Source ↗ |
| Automation scope | Recursion focuses on managed agents that perform ongoing work, use specialist agents, and produce rubric-graded results. | UiPath describes a broader automation platform that combines agents, robots, and people in business processes. Source ↗ | Power Automate supports cloud flows and desktop-flow automation, including attended desktop flows under its free license. Source ↗ |
| Work initiation | Recursion agents can start from schedules, events, Slack messages, webhooks, and API calls. | UiPath's public plan page describes building and running automations, but does not specify the same Recursion-style session-hour pricing or grading approach. Source ↗ | Power Automate licensing documentation covers cloud-flow and desktop-flow runs, with entitlements depending on license and flow context. Source ↗ |
Only factors supported by the available research are shown. The comparison is descriptive, not a ranking.
These products address broadly similar requirements. The descriptions focus on documented reasons a buyer might compare them with Labelbox.
| Company founded | 2018 |
|---|---|
| Headquarters | San Francisco, California, United States |
| Hosting | Managed infrastructure for Recursion agents |
| Pricing model | Agent-session hour, with compute and model spend estimated before launch |
| Built-in integrations | Slack, GitHub, Jira |
| Application connectivity | More than 100 business apps and MCP servers, according to Labelbox |
| Public introductory credit | $20 for new teams for a limited time |
| Security controls described | Isolated sandboxes, vaulted credentials, scoped permissions, approval gates, and session records |
Verified installation, hosting, administration, or environment details that can affect implementation.
| Hosting | Recursion Managed Agents are described as running in Labelbox-managed infrastructure and provisioned sandboxes. Source ↗ |
|---|---|
| Application access | Agents connect to supported business applications and can use a browser for tasks where an application has no API. Source ↗ |
A proprietary editorial score based on product capabilities, usability, value, performance, support and user sentiment evidence.
Current product capabilities and usage-based pricing are described on Labelbox's site, but no independent reviews of the newly launched Recursion Managed Agents were found. G2 feedback concerns the previous data-labeling product, so scores are cautious editorial assessments rather than user ratings or hands-on test results.
Methodology v1.0. The assessment weighs feature coverage (25%), ease of use (15%), value for money (20%), performance (15%), support (10%) and user sentiment (15%), using researched product evidence and verified review evidence where available. It is a TechZella editorial assessment, not a direct user-review average.
Key claims are linked to the sources used to verify them. This evidence record is separate from third-party user ratings.
TechZella evaluates software using documented product capabilities, pricing information, usability and value considerations, independent user evidence where available, and competitive context. The assessment is an editorial review, not a hands-on test unless one is explicitly identified. Our editorial assessment is based on 5 documented sources.
The company About page lists the founding year and San Francisco headquarters.
View source ↗Verified 2026-11-12The company describes Horizon as reinforcement-learning environments and evaluations for frontier models, and Recursion as managed enterprise agents.
View source ↗Verified 2026-11-12The launch page states support for 100+ business applications and MCP servers, and identifies Slack, GitHub, and Jira as built-in integrations.
View source ↗Verified 2026-11-12The product announcement says customers pay by the agent-session hour and see compute and model spend in an estimate before launch.
View source ↗Verified 2026-11-12The announcement describes controls for agent tools and credentials, human approvals for sensitive actions, and auditable session records.
View source ↗Verified 2026-11-12The November 11, 2026 announcement describes Wipro's work with the two products and its role helping customers connect systems and define workflows.
View source ↗Verified 2026-11-12The G2 product review page displays the aggregate rating and count, but those reviews are not specific to the newly launched Recursion agents.
View source ↗Verified 2026-11-12Answers begin with the clearest available response, followed by the context that helps you act on it.
No reviews yet. Be the first to share your experience.
Please log in to submit a review.