Harness is expanding its platform with Agent DLC, an approach that covers the entire lifecycle of AI agents. Developers can now build, test, deploy, manage, and govern agents using the same pipelines and controls they already use for regular software.
The idea behind Agent DLC is that an agent is simply software that goes through the same CI/CD pipeline as other services. New features include Harness AI Evals, which link quality gates to agents, and AI configs for managing prompts and model changes at runtime. Agent Deployments is also new, supporting managed runtimes such as Amazon Bedrock AgentCore. Harness had previously released a managed harness and CLI for this purpose.
Security at every stage
At Harness, security isn’t a standalone step; it’s woven into build, test, deploy, and operate. The new Harness Agent Security module combines shift-left and shield-right approaches. On the left side are Primitive Scanning, AIBOM (an extension of the classic SBOM), and AI Testing against the OWASP Top 10 for LLMs. On the right side, Agent Discovery and an AI Firewall protect against prompt injection and runtime data exfiltration. Harness had previously introduced two modules for securing AI code and AI apps.
Overarching this is AgentTrace, which provides end-to-end tracing and auditing for every execution. Harness is making its underlying components, including harness-sdk and harness-evals, open source. The platform also handles token cost management, an area for which Harness had previously launched two management platforms.