1. GenAI System Fundamentals

– Explain how foundation and reasoning models work, including tokens, context windows, and inference.
– Assess capability, latency, and cost trade-offs across model classes, including extended thinking.

2. Retrieval and Grounding Architecture

– Design retrieval-augmented generation (RAG) and grounding approaches for enterprise data.
– Evaluate data readiness for agents, including quality, permissions-aware retrieval, and knowledge lifecycle.

3. Context Engineering and Memory

– Define context strategies spanning system prompts, retrieved context, and tool results.
– Manage memory and compaction approaches for long-running agents and workflows.

4. Enterprise GenAI Reference
Architecture

– Sketch reference architectures covering channels, orchestration, models, tools, data, and guardrails.
– Identify observability, evaluation, and guardrail components required for production readiness.

5. Agentic System Design

– Distinguish prompted responses, workflows, single agents, and multi-agent systems — including when not to use agents.
– Design agent loops with goals, tools, policies, and bounded autonomy under the principle of least agency.

6. Workflow Control and Human
Oversight

– Design human-in-the-loop checkpoints and approval gates for agent actions.
– Plan failure handling for runaway loops, over-delegation, and cascading failures across connected agents.

7. Tool and Data Connectivity (MCP)

– Specify agent-to-tool and agent-to-data connectivity using the Model Context Protocol.
– Vet MCP servers and connectors as governed supply-chain components.

8. Agent Interoperability and Skills (A2A,
Agent Skills)

– Map cross-vendor agent collaboration using the Agent2Agent protocol.
– Package and govern procedural knowledge as portable agent skills.

9. Enterprise Platform Evaluation

– Compare the major enterprise agent ecosystems at a capability and governance level.
– Assess protocol support, exit costs, and vendor deprecation risk in platform selection.

10. Agent Security Management

– Threat-model agentic systems against leading agentic threat categories, including goal hijack and indirect prompt injection.
– Define mitigations for memory poisoning, tool compromise, and rogue-agent scenarios.

11. Agent Identity and Access
Management

– Specify authentication, least-privilege access, and lifecycle governance for agent identities.
– Design agent inventory, containment, and kill-switch patterns to manage agent sprawl.

12. Code-Centric and Computer-Use
Agents

– Evaluate agent harnesses, subagents, sandboxed execution, and browser or computer-use agents.
– Position code and computer-use agents appropriately within the SDLC and knowledge work.

13. Evaluation and Observability

– Define evaluation criteria before building, including quality, safety, and drift measures.
– Implement tracing, monitoring, and rollback practices for AI systems in production.

14. Emerging AI Technology Evaluation

– Stay informed of model, protocol, and platform developments.
– Assess the applicability and maturity of emerging AI capabilities for program goals.