Harness introduces Agent DLC for the AI agent lifecycle
The DevOps company is treating agents as governed software artifacts, with eval gates, canary rollouts, and tracing. It cites Gartner that just 8 percent of firms run agentic AI in production.
- Five products: AI Evals, Agent Deployments, AI Configs, AI Asset Catalog, and AgentTrace.
- Security in the loop: primitive scanning, an AIBOM, adversarial testing, and a runtime AI firewall.
- The gap: Harness cites Gartner that only 8 percent of organizations run agentic AI in production.
Harness is treating AI agents like any other piece of software that has to ship: built, tested, governed, and watched in production. On July 21 the DevOps company introduced Agent DLC, extending its platform to cover the full AI Agent Development Lifecycle so enterprises can build, deploy, and operate agents through the CI/CD pipelines and controls they already run.
Five products for the agent lifecycle
Agent DLC ships as five pieces. Harness AI Evals provides eval datasets, scoring functions, and quality gates to catch regressions before release. Agent Deployments brings canary releases and approvals to managed runtimes such as Amazon Bedrock. AI Configs lets teams change prompts and models at runtime through feature flags. The AI Asset Catalog auto-discovers agents, skills, and plugins. AgentTrace records multi-step agent runs and their execution paths. The throughline is familiar to any release engineer: version it, gate it, roll it out slowly, trace it when it breaks. The full breakdown is in the company's announcement.
Security moves into the loop
Harness paired the lifecycle tooling with a security layer: primitive scanning, an AI bill of materials it calls an AIBOM, adversarial testing, and a runtime AI firewall. CEO Jyoti Bansal framed the pitch as consolidation, saying teams can now apply "governance, orchestration, security, testing" to agents in one platform. An AIBOM is the agent-era answer to the software bill of materials that already tracks open-source components, a way to know what is actually running inside an autonomous system. The runtime firewall matters because agents, unlike static code, make decisions at execution time; adversarial testing and a firewall are how a team catches a prompt injection or a rogue tool call before it reaches production data. That security framing is what separates a demo from a deployment an auditor will sign off on.
The production gap
The urgency has a number behind it. Harness cites Gartner that only 8 percent of organizations have agentic AI in production, a gap it argues stems less from model capability than from missing operational scaffolding. As SiliconAngle reported, the bet is that developers adopt agents faster when they can use familiar processes and tools rather than a new stack. Agent DLC is rolling out now to existing customers. It sits alongside a wider push to make agent orchestration an operational discipline, and it mirrors hyperscaler moves like Tencent Cloud's AgentOps platform on the other side of the market.
What this means for the agent economy: the constraint on GaaS adoption is shifting from what agents can do to whether enterprises can audit, gate, and roll them back. Treating agents as governed software artifacts, with eval gates and canary rollouts, is the unglamorous work that lets autonomous systems reach regulated production at all.