Agentic SdLC

Expanded Definition

Agentic SdLC describes a software delivery lifecycle in which AI agents are granted bounded authority to perform engineering tasks such as opening pull requests, triaging tickets, generating code, running tests, and initiating reviews. The key distinction is not automation alone, but delegated execution with system access and decision-making steps that must be controlled, logged, and revocable.

Definitions vary across vendors and product teams, because some use the term for any AI-assisted development workflow while others reserve it for workflows where the agent can act without a human approving every step. For NHI Management Group, the security issue is the same either way: once an agent can touch repositories, CI/CD, ticketing, secrets, or deployment tooling, the lifecycle becomes an identity and control problem as much as a development process. The most relevant standards language today comes from governance and agentic ai risk sources such as the NIST AI Risk Management Framework and the OWASP Agentic AI Top 10, both of which emphasise accountability, bounded authority, and misuse resistance.

The most common misapplication is treating an agentic lifecycle as a productivity feature only, which occurs when teams give agents broad write access to code, tickets, or secrets without clear approval gates and audit trails.

Examples and Use Cases

Implementing Agentic SdLC rigorously often introduces workflow friction, requiring organisations to weigh delivery speed against tighter approval, logging, and rollback controls.

• An agent drafts a feature branch from a Jira ticket, but a human approves the merge after code review and test validation.
• An agent triages CI failures, updates the issue tracker, and suggests remediation steps while being denied direct access to production secrets.
• An agent creates dependency upgrade pull requests and runs test suites, with scoped repository permissions and signed audit logs.
• An agent generates release notes from commit history, but publication is gated by an approver who can verify accuracy and scope.
• An agent monitors backlog items and opens security tickets when it detects risky patterns, aligned to guidance from the CSA MAESTRO agentic AI threat modeling framework and the MITRE ATLAS adversarial AI threat matrix.

In more mature environments, engineering teams also use agentic workflows to support secure code remediation, but they segment privileges so that the agent can propose changes without independently altering sensitive infrastructure.

Why It Matters for Security Teams

Agentic SdLC matters because it changes the trust model of software delivery. Once an AI agent can act inside development and operational systems, standard developer tooling becomes part of the attack surface. Weakly governed agents can create unauthorized changes, leak secrets, approve unsafe dependencies, or amplify malicious instructions embedded in tickets, pull requests, or documents. That makes access scope, approval policy, and provenance controls central to the security design, not optional process overhead.

This is also where identity governance becomes practical. An agent that commits code or updates tickets is not just a model output generator, it is an actor with an operational identity that needs least privilege, traceability, and revocation paths. Security teams should align controls to the agent’s authority boundaries, then test how those boundaries behave under prompt injection, tool abuse, and compromised workflows. The issue is increasingly visible in public reporting, including the Anthropic first AI-orchestrated cyber espionage campaign report, which underscores how agentic tooling can be repurposed when controls are weak.

Organisations typically encounter the real impact only after an agent has made an unsafe change, exposed a secret, or polluted an audit trail, at which point Agentic SdLC becomes operationally unavoidable to address.

Related resources from NHI Mgmt Group

• What is Agentic AI and how does it differ from traditional generative AI?
• What NHI types do Agentic AI systems typically use?
• Why does Agentic AI dramatically increase identity sprawl?
• What is identity spoofing in Agentic AI and how does it work?