[DigitalToday reporter Chi-gyu Hwang (황치규)] Harness has launched 'Agent DLC' to help run AI agents within existing software development and deployment systems.

SiliconANGLE reported on July 21 that the product is designed to reduce issues that can arise when companies move agents from the concept stage into production by integrating agent-specific evaluation, deployment governance and security controls.

It supports developing and managing AI agents as they are within existing continuous integration and continuous deployment pipelines. Development teams can build, test, deploy and operate agents using the same platforms, control methods, pipelines and governance rules used for existing applications, without creating a separate system.

Harness said companies are rushing to adopt AI agents in hopes of improving productivity, but few cases have reached actual operations. Its internal data showed only 8 percent of organizations have deployed agentic AI in production. While general software is likely to repeat the same results once validated, AI agents are hard to predict because they decide on their own which APIs, tools and task sequences to use for each request.

Agent DLC provides specialized capabilities across the development lifecycle. 'Harness AI Evals' defines evaluation datasets and quality gates to check for performance degradation whenever an agent or model changes. 'Harness Agent Deployments' supports managed agent runtimes on external continuous deployment platforms, such as Amazon Bedrock AgentCore, to keep existing release procedures in place.

In operations, AI Configs supports releases, prompt management and in-run model changes. It can roll back immediately without redeploying a previous version. Harness' internal developer portal has also added an asset catalog that automatically discovers and registers agents, skills and plugins.

It also includes security features. The Agent Security module scans the AI models and skills used by agents to find configuration errors and tests adversarial inputs before deployment. It generates a bill of materials for each agent component and serves as a firewall in production to prevent prompt injection attacks and data leaks.