Introducing the Harness Agent DLC: New Abilities for the AI ​​Agent Development Lifecycle

SAN FRANCISCO – Harness, an AI Software Delivery Platform company, today announced that it is expanding its platform to include the complete AI Agent Development Lifecycle (DLC), giving businesses a single set of pipelines and controls to build, test, deploy, and run agents in the same way they already deploy everything else.

Every business is building AI agents, but most can’t find internal pilots or proofs of concept. According to Gartner®, “Only 8% of organizations have agent AI in production.” Enterprises’ reliance on the software delivery lifecycle through application code has not yet extended to agents, holding back the ROI of internal AI investments. The real innovation comes when a company can deploy an agent live with the same trust and confidence it has in all of its software.

“When we started Harness, the vision was a code security approach,” said Jyoti Bansal, co-founder and CEO of Harness. “Until recently, that meant application code. Today it also means agent code, written across engineering, product, sales, and support teams alike, building each agent to continue its work. Everything you’ve done for software delivery over the past decade — governance, orchestration, security, testing — you can now do for agents in the same place.”

Why AI agents are breaking the software delivery lifecycle

Native software works because it is predictable. The application code is decisive. Run the same test against the same code twice, and it produces the same result both times.

Agents don’t work that way: the underlying language model of the agent determines how it completes the task, and the same agent, given the same input, may choose a different tool or perform a different action from one to the next. Passing the test once does not guarantee that it will pass in the future. Incidents stop reoccurring when needed, which means the standard playbook for catching and fixing bugs doesn’t carry over either.

Rates increase with business size. A rogue agent could expose customer data, violate a compliance policy, or perform an unauthorized action. Businesses need a way to respond to what their agents are doing, and the traditional software delivery lifecycle wasn’t designed to provide that.

New Harness Agent DLC products and abilities

Agent DLC bridges the gap between developing an agent and delivering it safely into production. Today’s launch includes five new products and capabilities that include testing, deployment, operations, and management:

• Harness AI Evals make agent quality measurable, allowing teams to define eval datasets, connect scoring functions, and set quality gates that automatically catch regressions whenever an agent or model changes.

• Agent Deployment extends the canary release, authorization, and OPA directives that Harness already uses in Kubernetes deployments for managed agent runtimes such as Amazon Bedrock AgentCore and Google’s Agent Runtime. Agents now deploy through existing pipelines instead of cloud-specific workflows.

• AI Configs supports the release and management of notifications and model changes at runtime, supported by the same flagging feature infrastructure that already manages code release. Teams can test what works best and bounce back quickly, without rework.

• The AI ​​Asset Catalog automatically finds every agent, skill, and plugin built in every organization’s repositories and connects each one to an owner, so nothing goes or goes unnoticed.

• Harness AgentTrace records what happens during a single agent run and in a full multi-step session, showing which path the agent took, where it slowed down, and how different models or inputs affect the result. Harness also open source the basic components behind AgentTrace, including harness-sdk and harness-evals, so developers can bring the same trace elements to their AI applications.

In addition, existing Harness products already extend to agents without requiring any changes: Continuous Integration builds like any other service, Artifact Registry tracks its versions and dependencies, AI Test Automation verifies its answers in plain English, and AI Cost Management increases cost visibility across agents and models.

Protecting Agent DLC

Agents choose their own way and way to get there, so their behavior is hard to predict and hard to defend against. They expand their attack surface by connecting to tools and APIs, expose micro-agents, and gain trust from every model they touch. Solid scanners are never designed for this type of vulnerability. Harness introduces new security capabilities to close that gap.

Shift-left: limit what agents can do before they send.

• Primitive Scanning flags inaccuracies in agent capabilities, information, and models.

• AIBOM captures every model, tool, and dependency an agent is built with.

• AI testing uses agents against adversary input and OWASP Top 10 LLM and Agentic AI risks.

Shield-right: apply policy and keep visibility when they go live.

• Agent Discovery and Posture Management continues to map agents as they move and how they connect throughout the organization.

• AI Firewall implements policy in real time against rapid injection, misuse of tools, and data exfiltration.

Together, these capabilities give Agent DLC a single test path from development to production.

Built in Harness area

Harness creates context and intelligence locally through the Software Delivery Knowledge Graph, which captures and connects data from all stages of the delivery lifecycle, which now includes both applications and agents. Organizations that rely on slider tools don’t have that same connected view.

In June 2026, Harness introduced Autonomous Worker Agents, a platform for building and securely deploying AI agents within software delivery pipelines. Worker Agents act as controlled steps within those pipelines, covered by the same controls Harness already applies to all deployments.

Agent DLC extends that same context and governance throughout the agent’s lifecycle. Pipelines, policies, authorizations, and authentications that already work in the organization’s code now also work in its agents, so eval gates, deployment authorization, and security testing work as stages within a single pipeline, from the moment the agent is created to everything it does thereafter.

Availability

The Harness Agent DLC skills are rolling out now for Harness customers. For a full description of what is included in each stage of the life cycle, visit [blog page].

How does Harness AI Evals work in testing AI agents?

Harness AI Evals allows teams to define datasets for analysis, configure scoring functions, and set quality gates that automatically detect regressions when the agent or underlying model changes. It addresses the nondeterministic nature of LLM-based agents, where the same input can produce different outputs across runs.

Can the Harness Agent DLC deploy agents to Amazon Bedrock AgentCore and Google Agent Runtime?

Yes. Agent deployment capabilities extend Harness canary output, authorization workflows, and OPA policy directives to managed agent runtimes including Amazon Bedrock AgentCore and Google’s Agent Runtime, so agents deploy through existing Harness pipelines rather than cloud provider-specific workflows.

How does Harness protect AI agents against rapid injection and LLM vulnerabilities?

The harness offers both left-handed and right-handed safety capabilities. Shift-Left’s tools include Core Scanning for misconfigurations, AIBOM for dependency tracking, and AI testing against OWASP Top 10 LLM and Agenttic AI risks. Shield-right’s capabilities include Agent Discovery and Configuration Management for runtime visibility and an AI Firewall that enforces policy against rapid injection, tool misuse, and real-time data exfiltration.