AI applied to agile work

Holding these up-to-date diplomas certifies that the holder is able to:

1) Manage the context the model receives

Distinguish context engineering from prompting, recognize when a task calls for managing context, justify why adding information can worsen the result, and apply the minimum sufficient context, with what is decisive well placed and the write, select, compress, and isolate strategies.

2) Ask with structure and reuse what works

Build prompts with an objective, success criteria, context, format, and constraints, diagnose which component is missing when they fail, apply the providers' current recommendations, and turn effective prompts into team templates.

3) Verify before acting without losing one's own judgment

Separate facts, inferences, and recommendations, ask for the assumptions and evidence behind a claim, treat any figure without a source as suspect, calibrate verification to the risk, and use AI in a way that preserves one's own understanding when learning something new.

4) Protect the trust boundary

Treat external content as untrusted, evaluate an agent's configuration with the lethal trifecta or the Agents Rule of Two, recognize the roots of excessive agency, and apply data minimization and effective anonymization before sharing information with a model.

5) Agree on and maintain the team's AI usage policy

Explain why shadow AI appears, classify uses as prohibited, allowed with guardrails, and allowed, write a one-page policy and build it into the retrospective, the Definition of Done, and the onboarding of new people, knowing the AI literacy obligation.

6) Delegate verifiable work to agents

Distinguish an assistant from an agent, decide whether a task can be delegated based on its verifiability and the reversibility of its consequences, reserve for human judgment what requires it, frame the delegation with limits and acceptance criteria, and anticipate its impact on the team's events and artifacts.

7) Design human oversight

Choose between human-in-the-loop, human-on-the-loop, and human-out-of-the-loop according to risk, place approval gates, recognize nominal oversight and automation bias, design a traceability record, and explain the obligations of the EU AI Act and its current timeline.

8) Understand the agentic ecosystem standards

Explain what MCP, AGENTS.md, A2A, and Agent Skills solve and how they differ, interpret their neutral governance as a sign of consolidation, and identify the risks of connecting third-party servers or skills.

9) Apply AI to the backlog with the decision in the team

Use AI as a reviewer of stories with explicit criteria, generate refinement questions, request and assess slicing proposals, write acceptance criteria with negative scenarios and edge cases, and frame experiments with metrics and guardrails.

10) Support facilitation without replacing human interaction

Prepare sessions and synthesize their material with AI while contrasting the synthesis with one's own, recognize the boundary of what is not delegated, explain the ban on inferring emotions at work, and record or transcribe sessions with the proper safeguards.

11) Use AI in quality and delivery with technical rigor

Interpret the evidence on the security of generated code and on perceived productivity, generate tests that target the real risks and require an independent origin for them, review code with AI support while keeping the decision, and recognize hallucinated dependencies and injection in the development cycle.