iSAQB CPSA-A AGENTA

Beschreibung

Attending the iSAQB® CPSA-A Architecture for Agentic Software Engineering Contexts (AGENTA) course gives participants 20 Methodological Competence (MC) points and 10 Technical Competence (TC) points towards the 70 points required for eligibility for the iSAQB CPSA-A exam with Brightest. It is important to remember that, as part of the 70 points required to take the iSAQB CPSA-A exam with Brightest, you will need at least ten competence points in each of the following areas: 

• Technical Competence (TC)
• Methodological Competence (MC)
• Communicative Competence (CC)

Accredited iSAQB® AGENTA - Architecture for Agentic Software Engineering Contexts (CPSA-A) training is based on the current iSAQB® curriculum:

Part 1 - Introduction and Fundamentals

• Understand the relationships between artificial intelligence, machine learning, generative AI, and large language models
• Distinguish between AI-assisted, augmented, and agentic coding
• Understand AI orchestration, agentic workflows, and agentic engineering
• Recognize the differences between agentic engineering and vibe coding
• Understand the capabilities and limitations of large language models, including hallucinations, bias, knowledge cut-offs, and context limitations
• Explain the influence of prompt engineering and context engineering on the quality of AI-generated results
• Distinguish between AI assistants and AI agents
• Understand agentic loops, tool usage, state management, planning, sub-agents, and human-in-the-loop approaches
• Classify different levels of autonomy in agentic engineering
• Know approaches such as Spec-Driven Development, agent swarms, and conversational programming
• Decide whether a task should be performed by a human, a traditional software tool, or an AI agent
• Understand how AI assistants and agents can support software architecture activities while preserving human architectural judgment

Part 2 - AI-Assisted Architectural Decision-Making

• Understand the lifecycle of architectural decisions and how agentic engineering affects it
• Establish continuous feedback loops between architecture, development, testing, and operations
• Use AI to support continuous, evidence-based architectural decision-making
• Refine quality goals into assessable architectural drivers and quality scenarios
• Identify conflicts between quality goals, constraints, and architectural intentions
• Identify structural weaknesses, architectural drift, technical debt, and emerging complexity
• Combine established analysis tools with AI-assisted architecture reviews and assessments
• Use AI to generate, refine, and compare architectural alternatives
• Evaluate architectural options against drivers, constraints, principles, and existing decisions
• Treat architectural decisions as testable hypotheses with explicit assumptions and validation criteria
• Design lightweight experiments, prototypes, simulations, spikes, and side-by-side comparisons
• Recognize the limitations and risks of AI-supported architectural reasoning
• Define review checkpoints, guardrails, and boundaries for consequential architectural decisions
• Shift architectural work from prescribing implementation details towards orchestrating goals, constraints, and acceptance criteria
• Recognize and contain orchestration risks such as runaway agents, infinite loops, and cascading errors
• Consider cost and sustainability when designing agentic workflows

Part 3 - Providing Architectural Knowledge to Agents

• Understand how the information available to an AI agent influences the quality and architectural consistency of its output
• Treat the provision of architectural knowledge to agents as an engineering and design challenge
• Identify the architectural decisions, constraints, quality goals, principles, domain concepts, and system boundaries relevant to AI agents
• Design precise, maintainable, and effective agent-facing representations of architectural knowledge
• Use formats such as Architecture Decision Records, architecture rules, context files, knowledge bases, retrieval indexes, diagrams-as-code, and structured constraints
• Select suitable mechanisms for delivering architectural knowledge to agents at the right time
• Assess strategies such as Retrieval-Augmented Generation and reusable agent skills
• Understand the architectural implications of standardized agent-tool integration protocols such as the Model Context Protocol
• Evaluate whether the architectural knowledge available to agents is sufficient and appropriately structured
• Balance comprehensive architectural context against the risk of irrelevant or contradictory information
• Improve knowledge delivery iteratively by observing and evaluating agent behaviour
• Recognize that shared documentation does not automatically create shared understanding

Part 4 - Keeping Agents Aligned with Architectural Goals

• Understand why AI agents require explicit architectural controls and constraints
• Recognize how unconstrained AI-generated work can introduce inconsistencies, technical debt, and architectural drift
• Understand why architectural knowledge alone is insufficient to control agent behaviour
• Design preventive and detective controls that keep AI-generated work within architectural boundaries
• Assess trade-offs between precision and coverage as well as speed and depth of analysis
• Select suitable enforcement mechanisms for structural rules and broader quality characteristics
• Use architecture tests, static analysis, linters, dependency rules, policy checks, review gates, sandboxing, and approval workflows
• Position architectural controls at appropriate points in the development and delivery process
• Establish continuous architectural verification for rapidly evolving AI-assisted codebases
• Assess whether an existing codebase and technology stack are suitable for architectural control of AI agents
• Introduce architectural controls incrementally into existing systems
• Balance the effort required to establish controls against the risks of operating agents without them
• Use failures in AI-generated work to improve architectural controls over time
• Determine where control improvements can be automated and where human judgment remains essential

Part 5 - Architecture Information Extraction

• Distinguish between as-is and to-be architecture documentation
• Distinguish between human-facing and agent-facing architectural information
• Understand where AI can support architectural documentation and where human collaboration remains necessary
• Create concise and machine-readable representations for AI-assisted workflows
• Maintain architectural decisions and documentation throughout agentic development loops
• Detect when agentic development implicitly changes or moves away from existing architectural decisions
• Use Architecture Decision Records to capture decisions, alternatives, rationale, and consequences
• Use AI to draft ADRs from context logs, guardrail outputs, conversations, emails, and meeting records
• Establish human review checkpoints to preserve traceability and accountability
• Extract architectural information from source code, repositories, interfaces, configuration, tests, and deployment descriptors
• Combine AI-generated insights with static and dynamic analysis
• Detect technologies, frameworks, architectural patterns, design patterns, and candidate business rules
• Identify gaps, inconsistencies, and uncertainties in recovered architectural information
• Generate and maintain architectural views from development artifacts
• Use text-based formats such as Mermaid and Structurizr for version-controlled living documentation
• Validate generated documentation against developer knowledge and actual system behaviour

Part 6 - Governance and Quality Gates for AI Usage

• Understand copyright and software-licensing considerations when using AI-generated code and artifacts
• Establish controls that reduce copyright-related risks in AI-supported development
• Assess the processing of personal and sensitive data when using AI tools
• Understand roles and responsibilities under data-protection requirements
• Apply the basic roles, risk logic, and phased requirements of the EU AI Act to typical AI-tool scenarios
• Identify when data-protection, information-security, legal, and other stakeholders should be involved
• Document AI-related decisions in an audit-ready and decision-ready manner
• Ensure the traceability and provenance of AI-generated outputs
• Use structured logging, context references, citations, versioning, tracing, and monitoring
• Scale AI governance consistently across teams and development processes
• Compare token-based and subscription-based AI cost models
• Consider vendor lock-in, provider flexibility, and data sovereignty when selecting AI services
• Use guardrails, approval processes, and allowlists for agent and tool integrations
• Understand the environmental impact of AI-supported and agentic development
• Develop sustainable usage strategies based on model size, workflow design, and resource consumption
• Define secure and data-protection-compliant rules for selecting and using AI tools
• Understand security risks such as prompt injection, tool misuse, and confused-deputy attacks
• Apply measures such as least privilege, scoped credentials, explicit approvals, and monitoring
• Select suitable container-based or virtual-machine-based sandboxing strategies

Part 7 - The Evolving Discipline of Software Architecture

• Maintain responsibility and ownership for AI-generated code and engineering outcomes
• Recognize and address the risk of engineering skill loss through overreliance on AI
• Use deliberate practice, human-first reasoning, and AI-supported learning to preserve core skills
• Apply critical thinking when reviewing and validating AI-generated results
• Distinguish between AI-supported reasoning and delegated decision-making
• Identify architectural activities that must remain under human ownership
• Use shared terminology and domain vocabularies to improve the consistency of agent decisions
• Measure whether teams are improving architectural capability rather than merely producing outputs faster
• Adopt AI-supported development through explicit hypotheses, controlled experiments, and measurable outcomes
• Understand AI adoption as an ongoing process of access, adoption, proficiency, changed working practices, and organizational development
• Support adoption through training, experimentation, clear ownership, and psychological safety
• Use meaningful indicators such as lead time, change failure rate, DORA metrics, and developer-experience signals
• Apply collaboration patterns that combine humans and AI while maintaining effective communication between team members
• Understand agent topologies such as solo agents, sub-agents, and agent teams
• Adapt roles, team structures, engineering practices, and architectural responsibilities to AI-supported development
• Shift from line-level implementation towards broader architectural, product-oriented, and outcome-focused thinking
• Understand how AI affects software, solution, and enterprise architecture