Definition
CI/CD (Continuous Integration / Continuous Delivery) encompasses automated build, test, and deployment pipelines that validate code changes before release. Traditionally deterministic and reproducible; in 2026 evolving to incorporate agentic automation alongside conventional workflows.
GitHub Actions Dominance
github Actions is the primary CI/CD platform for millions of repositories. June 2026 developments:
- Agentic Workflows public preview (June 11): Natural-language markdown compiles to Actions YAML; coding agents handle triage, CI failure analysis, docs updates (2026-06-11-github-agentic-workflows-public-preview)
- GITHUB_TOKEN support: No PAT required for agentic workflows with
copilot-requests: writepermission (2026-06-11-github-agentic-workflows-github-token) - New runner images: Ubuntu 26.04, Windows 11 arm64 (public preview)
Agentic vs Deterministic CI/CD
GitHub documentation explicitly states agentic workflows complement — not replace — core CI/CD. Build and release processes requiring strict reproducibility should remain conventional Actions workflows.
| Aspect | Deterministic CI/CD | Agentic Workflows |
|---|---|---|
| Behavior | Fixed scripts, reproducible | Reasoning-based, adaptive |
| Use case | Build, test, deploy | Triage, analysis, docs, review |
| Security model | Secrets in jobs | No secrets in agent runtime; safe outputs gate |
Security Layers (Agentic)
- Read-only permissions by default
- Agent Workflow Firewall + sandboxed containers
- Safe outputs validation
- Threat detection on proposed changes
- Bot PR approval gates for workflow execution
Model ID in CI/CD Pipelines
anthropic Claude API model retirements (June 15, 2026) affect CI/CD configs with hardcoded model strings in GitHub Actions, Docker, and Kubernetes. Audit claude-sonnet-4-20250514 and claude-opus-4-20250514 before cutoff (2026-06-12-anthropic-claude-api-june-15-deprecation).
Supply Chain Context
- miasma-worm campaign (June 2026) targeted GitHub Actions secrets and CI runners
- Agentic CI expands attack surface if external telemetry trusted (agentjacking-ai-coding-agents-sentry)