Recursion Managed Agents

AI agent · Workflow / back-office agent · Operations & supply chain by Labelbox

Managed AI agents for engineering and research teams that get better every run.

Recursion Managed Agents is an AI agent platform for enterprises to automate long-horizon engineering and research tasks. It runs fleets of specialist agents that fix bugs, investigate incidents, and run experiments across 100+ integrated apps. Choose it to clear technical backlogs with autonomous agents that learn from every graded run.

Agent specs

The full spec card (context, pricing per token, modalities and more) on AntSpace. Sign up free to see the full spec card.

Languages
English

Category: AI Agents

Languages
English (implied by interface and documentation)
Architecture
Coordinator-Specialist Multi-agent
Execution Mode
Event-driven (Schedules, Alerts, API calls)
Learning Mechanism
Memory runs and model fine-tuning
Quality Control
Automated rubric-based grading per session
Environment
Managed sandboxes and credential handling
  • 100+ integrations across code, CI, incidents, data, and docs
  • Autonomous agents run 168 hours a week on schedules or events
  • Multi-agent architecture with coordinator and specialist agents
  • Automated grading of every session against custom rubrics
  • Continuous learning from graded sessions to improve over time

Overview

Recursion Managed Agents is a platform for running fleets of autonomous AI agents designed to handle complex engineering and research workflows. The platform uses a multi-agent architecture where a coordinator plans the work and spawns specialist agents to execute tasks like bug fixing, incident investigation, and experiment sweeps. Every session is automatically graded against custom rubrics by a separate grader agent, and successful outcomes are used to update the agent's memory. Recursion integrates with over 100 business applications, including Salesforce, Zendesk, Jira, and Slack, allowing agents to take real actions within existing toolchains. By running on schedules, alerts, or API calls, it automates repetitive technical work that typically requires manual intervention.

Agent Fleet Management

  • Multi-agent Architecture: A coordinator agent breaks down each request and spawns specialist agents in parallel to handle complex jobs.
  • Persistent Fleet Operations: Agents operate across repositories, CI/CD pipelines, clouds, and data systems to complete long-horizon tasks.
  • Long-horizon Execution: Sessions can run for days or weeks, pausing on dependencies and picking up exactly where they left off.
  • Event-driven Triggers: Agents start automatically based on schedules, failing builds, PagerDuty alerts, merged pull requests, or API calls.

Quality and Learning

  • Automated Outcome Grading: A separate grader agent evaluates every session against a custom rubric, criterion by criterion, using gathered evidence.
  • Continuous Learning Loop: Every finished session feeds the agent's memory, allowing them to get better with every run.
  • Memory Verification: Memory runs verify what sessions teach and update the agent's memory before the next session acts.
  • Fleet Visibility Dashboard: Users can see every agent at work in one place to monitor progress and quality across the entire fleet.

Specialist Agent Roles

  • Bug Fixer Agent: Reproduces new bugs, finds the root cause, and writes a fix with a corresponding regression test.
  • Incident Investigation Agent_: Correlates alerts, logs, and recent deploys the moment an alert fires to provide a root-cause draft.
  • Experiment Sweep Agent: Runs dozens of experiment configurations at once and compares every run against a baseline.
  • Feature Builder Agent: Builds ready tickets into working features with tests in isolated sandboxes.

Use cases

  • Engineering teams use Bug Fixer agents to reproduce new bugs and write fixes with regression tests automatically.
  • Reliability engineers deploy Incident Investigation agents to correlate alerts and logs the moment a PagerDuty alert fires.
  • ML researchers use Experiment Sweep agents to run dozens of configurations at once and compare them against baselines.
  • DevOps teams use Dependency Upgrader agents to keep packages current and fix breaking changes in CI/CD pipelines.

Specifications

Integrations
Salesforce, Zendesk, NetSuite, Slack, Jira, GitHub, PagerDuty, Datadog, LaunchDarkly, Stripe, Statuspage
Languages
English (implied by interface and documentation)
Architecture
Coordinator-Specialist Multi-agent
Execution Mode
Event-driven (Schedules, Alerts, API calls)
Learning Mechanism
Memory runs and model fine-tuning
Quality Control
Automated rubric-based grading per session
Environment
Managed sandboxes and credential handling

How it works

  1. Connect your stack: Connect Recursion to your existing stack using 100+ integrations including GitHub, Jira, and Datadog.
  2. Define agent goals: Define goals for your agents, such as fixing regressions, upgrading dependencies, or running experiment sweeps.
  3. Automated execution and grading: A coordinator agent plans the work, spawns specialists, and a grader agent verifies the outcome against your rubric.
  4. Continuous improvement: Every finished session feeds the agent's memory, allowing the fleet to improve performance with every run.

FAQ

How do agents start their work?

Agents can start based on a schedule, a message, an API call, a failing build, an alert, or a merged pull request.

How does the platform ensure the quality of agent work?

A coordinator agent plans the work and spawns specialist agents in parallel. A separate grader agent then evaluates the result against your specific rubric to ensure quality.

What apps does Recursion integrate with?

The platform features 100+ integrations across code, CI, incidents, data, and documentation tools, including Salesforce, Zendesk, NetSuite, Slack, and Jira.

Get started

Added by AntSpace from public sources (labelbox.com).

Related

More AI agents by Labelbox