SAN FRANCISCO – Harness, an AI Software Delivery Platform company, today launched Autonomous Worker Agents for software delivery: a platform for businesses to build and securely deploy AI agents that manage the work between writing code and sending it to production.
The delivery of the software went through phases. First, people did the work with their hands. Then they write scripts for each activity such as deployment. More recently, they’ve connected those jobs to automated pipelines that follow consistent instructions, which is what Harness has done for big businesses for years. Worker Agents are the next category. Every operational step can now act as a thinking agent instead of the dense, contextual script, governance, sandboxing, and test leads that businesses need to rely on for production agents.
A number of Harness Managed Agents are available today, and any team can customize them or build their own. Harness Agent’s new marketplace makes it easy to find, use, and share.
“AI now writes the code. Harness sends it,” said Jyoti Bansal, Founder and CEO of Harness. “Independent Worker Agents is how businesses build and securely run AI in everything behind the code: build, test, secure, deploy, operate. Everything works on the same lines that already deploy our customers’ software, within the perimeter of their network. Governance, audit trail, and security posture are already in place. Worker Agents inherit everything from day one”.
Controls that keep agents safe in production
Autonomous Worker Agents work with infrastructure controlled by the customer. Code and data never leave the customer’s network, and agents are governed by the same controls that businesses already use for human deployments.
These controls make Autonomous Worker Agents safe to run in production:
Sandboxing: Agents run in isolated containers with limited file and network access. The agent that generates the malicious command has no place to send the data.
Limited authentication: Each agent has its own identity and a specific set of permissions assigned to it, the same way an employee does. An agent can only take actions that those permissions allow, regardless of who is causing them or what your information says.
Policy maintenance: Uniform policies that include deployment gateway agents. Policy can keep agents out of unauthorized models or out of production pipelines.
Audit methods: All actions of the agent are recorded under a different identity of the AI, with the full name: what caused the agent, what it did, and the result.
Cost tracking: Token consumption is visible per agent, per pipeline, with a budget limit that stops an agent when it reaches its limit.
Chaining: Agents chaining a multi-step workflow, passing the output from one to the next.
Easy to build, governed by your policies
Building a Workflow Standalone Agent uses the same agent file format that has become standard across the industry. Save it to a single file, commit it to your repository, and the agent is live, managed, and available throughout your organization. Teams that prefer not to write a file can use Harness AI to generate an agent for them. In any case, the agent acts as a controlled process with the same controls, audit trail, and policy implementation as everything else.
Once started, the agent has the full context of your organization. It uses the Harness Software Delivery Knowledge Graph, a connected map of your services, pipelines, deployments, infrastructure, incidents, and security findings. The vulnerability assessment agent knows which services are affected and who owns them. A default agent knows which services depend on which one is being used. The result is an answer built for your specific environment, not a generic fix that just looks right.
Agents also meet you where you work. With the Harness MCP server, a developer in Cursor, Code Claude, or another tool can assign a Worker Agent a task and execute it in the Harness, with the result returned from wherever it was configured. Wherever the agent operates, it operates under your organization’s policies, governed in the same way as all other steps in your route.
The tools are made of harnesses, ready for use today
Engage dozens of Worker Agents who handle repetitive, time-consuming work that slows teams down throughout the delivery lifecycle. Here are a few agents that are available today, and are being added regularly:
Autoremediation reads the build logs, identifies the root cause of the build failure, makes the correction to the PR branch, and retriggers the build until it passes.
Code Review Review PR differs in code quality, security issues, and test coverage.
The Compiler code identifies the untested lines and generates tests to cover the gaps.
Java, React, and Python Library Development checks for dependencies and recommends safe development.
Feature Flag Cleanup finds old flags and ensures safe removal.
Manifest Remediator analyzes failed Kubernetes deployments and fixes manifest issues.
Zero Day Mitigation isolates vulnerabilities behind feature flags and generates candidate patches for public review.
Harness Agent Marketplace
The Harness Agent Marketplace is a shared catalog where Worker Agents are published and recycled across the organization and the wider Harness community. Teams can use an existing agent instead of building their own, and contribute the agents they build back to the catalog.
It has three sections:
Harness Managed: Built, maintained, and SLA supported by Harness.
Harness Certified: Developed by our partners, reviewed and verified by engineering and security teams.
Community: Published by the wider Harness community. Organizations can use out-of-the-box policies to control which public agents run in production.
Every agent on the Market can be plugged in. A team can customize an existing agent and adjust information, tools, or risks to fit their environment. The one agent team that builds to solve a problem becomes the starting point for the next team that hits the same roadblock.
Autonomous Worker Agents work with any LLM provider. Connect Anthropic, AWS Bedrock, or Google Vertex AI through existing Harness connectors, and change models per agent, per location, or per pipeline without rewriting the agent.
Availability
Self Employed Agents and the Harness Agent Marketplace are now available to all Harness customers. For more information, visit https://harness.io/platform/
