4lbert Logo~/4lbert
HomeProfileSkillsPortfolioBlogGalleryContact
$ echo "© 2026 Torsten Albert · 4lbert.com"
ImpressumDatenschutz
Back to blog

DevAgentOps: Workflow for LLM Agents in CI/CD

TorstenMarch 21, 20262 min read

DevAgentOps runs LLM agents inside CI/CD with hard gates: analyze role-based use cases/stories, write testable acceptance criteria and verify for contradictions/gaps; design by mapping features to modules, planning, modeling (formal checks where possible), refining to implementation-ready specs, and creating reverse prompts; implement via human-piloted agent coding plus agent review, MR, unit/integration/E2E/role tests, then manual merge - validate behavior and loop back with evidence

On this page
  • 1) Analyse (Define + Verify Requirements)
  • 2) Design (Plan + Prove Consistency)
  • 3) Implementation (Build + Review + Test)
  • 4) Validation (Confirm Real Behavior)

DevAgentOps is the operational practice of using LLM agents inside the software delivery pipeline with explicit requirements, verifiable design artifacts, controlled implementation, and tight feedback loops.

1) Analyse (Define + Verify Requirements)

  • Role-based use cases
  • Role-based story writing
  • Acceptance criteria creation (testable, unambiguous)
  • Acceptance criteria verification: detect contradictions, errors, missing details, and incompleteness/gaps; refine until consistent.

Exit: criteria are implementable and testable without interpretation.

2) Design (Plan + Prove Consistency)

  • From feature list + module linking, identify involved modules and interfaces.
  • Implementation planning (sequencing, dependencies, risks).
  • Modeling + (formal) verification of key flows/contracts/invariants.
  • Manual refinement to “implementation-ready” specs.
  • Reverse prompt creation derived from design + criteria (constraints, contracts, Do/Don’t rules).

Exit: design artifacts are consistent and sufficient to implement.

3) Implementation (Build + Review + Test)

  • Manual piloting of LLM-assisted implementation until it matches design/criteria.
  • Agent reviews (correctness vs. criteria, consistency, test gaps) until accepted.
  • Merge request creation with traceability (story → criteria → code → tests).
  • Testing:
    • Unit + integration tests
    • UI end-to-end tests
    • Role-based testing of use cases
  • Manual merge as the final control gate.

Exit: tests pass and acceptance criteria are demonstrably satisfied.

4) Validation (Confirm Real Behavior)

  • Validate: Is functional behavior correct?
  • If not: capture evidence and throw back to Implementation (and update tests; update criteria only if requirements were incomplete).

Exit: deployed behavior matches acceptance criteria.

PreviousBeyond Agile: Agentic Safety-Critical DevelopmentNextComplexity of Composition Problem in AI Coding Agents

On this page

  • 1) Analyse (Define + Verify Requirements)
  • 2) Design (Plan + Prove Consistency)
  • 3) Implementation (Build + Review + Test)
  • 4) Validation (Confirm Real Behavior)