Labelbox

Labelbox

4.1

Labelbox builds AI training environments and Recursion, an agent platform that handles recurring software, reliability, security and ML research work autonomously.

Follow
Tags
AI AgentArtificial Intelligenceai-workflow
Last Updated Oct 21, 2026
Labelbox screenshot

About

Labelbox is a company that builds environments used to train artificial intelligence, along with Recursion, a platform where autonomous software agents take on recurring work for engineering and research teams. The product addresses a common problem in technical organizations: high-value staff spend large portions of their time on repetitive tasks such as fixing flaky tests, upgrading dependencies, investigating incidents, patching vulnerabilities, and rerunning model evaluations.

Recursion agents do not wait to be prompted from inside a code editor. They start on their own in response to triggers such as a failing build, an alert from a monitoring tool, a new model checkpoint, or a set schedule, and they keep working in the background until the task is complete. Each session is graded by a grader agent, and graded sessions leave behind memories, notes, and skills that later sessions read before acting, so results improve over time. Labelbox states that frontier AI labs train on its environments and data, and the company has announced a Series E raise of more than $100 million.

Key Features

Autonomous agent sessions: A coordinator agent and multiple specialist agents work together on tasks such as tracing a model regression, rerunning evaluations, diffing training data, and bisecting code, then send the result for human review.
Software engineering agents: Feature Builder turns ready tickets into working features with tests, Bug Fixer reproduces and fixes new bugs with regression tests, and Dependency Upgrader keeps packages current.
Reliability and infrastructure agents: Incident Investigation correlates alerts, logs, and deploys when an alert fires; Cloud Cost Watch traces spend spikes to the change behind them; Postmortem Writer drafts timelines and follow-ups.
Security engineering agents: Code Security Review reviews pull requests and reproduces real vulnerabilities, Vulnerability Patcher patches services affected by new critical CVEs, and Access Review flags stale or risky permissions.
ML research agents: Experiment Sweep runs many configurations at once against a baseline, Model Trainer launches fine-tuning runs and stops bad runs early, and Ablation Runner identifies which changes actually mattered.
Graded outcomes: Every agent run receives a score from a grader agent, and each graded session records memories and skills that subsequent sessions use.
Event-driven triggers: Agents begin work from schedules, alerts, commits, and API calls rather than only from direct user prompts.

Pricing & Fees

  • Pricing: Not publicly listed on the pages we reviewed; check the official site for current plans.
  • Integrations: The site shows Recursion working with GitHub, Jira, Slack, Linear, Google Cloud, BigQuery, Databricks, Sentry, PagerDuty, Datadog, Okta, Vercel, Jenkins, Confluence, and Google Drive.
  • Trial availability: The site offers a "Try Recursion" option, suggesting direct access for interested teams.
  • Funding: Labelbox has announced a Series E round of more than $100 million.

Who It's For

Labelbox with Recursion is aimed at engineering and research organizations, particularly ML research teams, software engineering groups, security teams, and platform or reliability engineers at enterprises and frontier AI labs. Typical use cases include handing over recurring maintenance work such as fixing flaky tests, keeping dependencies current, investigating incidents, patching vulnerabilities, running experiment sweeps, and diagnosing model evaluation regressions, so that human engineers can focus on higher-value problems.

FAQ

What is Labelbox?Labelbox builds the environments that AI labs use for training, and Recursion, a platform of autonomous agents that complete recurring engineering, infrastructure, security, and ML research work in the background.

How much does Labelbox cost?Pricing is not publicly listed on the pages reviewed; interested teams should check the official Labelbox website for current plans.

How do Recursion agents get started?Agents start automatically in response to triggers such as failed CI runs, monitoring alerts, new model checkpoints, commits, schedules, or API calls, and they continue until the job is finished.

Which tools does Recursion integrate with?Recursion works with tools including GitHub, Jira, Slack, Linear, Databricks, BigQuery, Sentry, PagerDuty, Datadog, Okta, Vercel, and Google Cloud services, according to the vendor's site.

How does Recursion improve over time?Every session is graded by a grader agent, and graded sessions leave memories, notes, and skills that later sessions read before acting, which the vendor says makes each subsequent run better.