T4: AI Engineer
Track: AI Engineer
Agentic stack planes: Control, Runtime, Integrations, Security, Observability
Enterprise planes: Governance & Risk, Strategy & Portfolio, Economics & ROI
Target Audience: Engineers and ML engineers
Badge: Applied AI Engineer
Meaning: Governed autonomy builder
Agentic plane colors match the AI Capabilities Stack (round dots on track cards).
Enterprise plane colors match the same diagram (square markers on track cards).
Who This Is For
- Software engineers building AI systems
- ML engineers and data scientists
- AI system architects
- Developers working with LLMs and agents
What You'll Learn
Core Capabilities
- Agent Patterns - Design and implement AI agents effectively
- Multi-Agent Orchestration - Coordinate multiple agents in complex systems
- Guardrails & Safety - Build safety mechanisms into autonomous systems
- Governed Autonomy - Balance autonomy with control and oversight
Key Focus Areas
- Agent architecture patterns
- Control plane, tool registry (MCP), and protocols
- Safety guardrails and zero-trust identity
- Restricted autonomy (for public sector)
Schedule
Duration: 10 weeks
Weeks: 1-10
Focus: Production agentic systems - control through observability
Week 1: Architecture Overview
Bloom Level: Remember & Understand (primary), Apply (secondary)
Knowledge Dimension: Factual & Conceptual
Tool Mediation: Diagramming and reference architectures
- Agentic stack reference model and plane boundaries [Remember]
- Front door, workers, sandboxes, and inference lanes [Understand]
- Failure modes and blast radius [Analyze]
- Sketching your target architecture [Apply]
Cognitive Integration: Sets the map for the rest of the track - Transfer to your stack.
Week 2: Control Plane + Tool Registry
Bloom Level: Apply (primary), Analyze (secondary), Create (foundation)
Knowledge Dimension: Procedural & Conceptual
Tool Mediation: MCP servers, tool registry patterns, catalogs
- Tool registry design and MCP integration [Apply]
- Allowlists, versioning, and deprecation [Analyze]
- Deterministic policy vs prompt-only “hope” [Evaluate]
- Wiring tools safely for agents [Create]
Cognitive Integration: Governance and Tool Mediation at the control layer.
Week 3: Policy + Guardrails
Bloom Level: Apply (primary), Create (secondary), Analyze (foundation)
Knowledge Dimension: Procedural & Metacognitive
Tool Mediation: Policy engines, validators, and LLM guardrail SDKs
- Policy pipelines for inputs and outputs [Create]
- Structured checks vs model-only moderation [Analyze]
- Testing guardrails in CI [Apply]
- Escalation when policies conflict [Evaluate]
Cognitive Integration: Core Governance engineering.
Week 4: Identity + Zero-Trust
Bloom Level: Apply (primary), Understand (secondary), Analyze (foundation)
Knowledge Dimension: Procedural & Conceptual
Tool Mediation: OAuth/OIDC, workload identity, secrets
- Service-to-service and user-to-agent identity [Apply]
- Zero-trust patterns for tools and data [Understand]
- Secrets, rotation, and least privilege [Apply]
- Threat modeling for agentic flows [Analyze]
Cognitive Integration: Security plane foundations for engineers.
Week 5: Durable Orchestration + HITL
Bloom Level: Create (primary), Apply (secondary), Analyze (foundation)
Knowledge Dimension: Procedural & Conceptual
Tool Mediation: Workflow engines, durable execution platforms
- Long-running workflows and compensation [Create]
- Human approvals in the loop [Apply]
- Idempotency and replay [Analyze]
- Choosing orchestration vs ad-hoc scripts [Evaluate]
Cognitive Integration: Bridges runtime and control for reliable agents.
Week 6: Runtime Plane Engineering
Bloom Level: Apply (primary), Create (secondary), Evaluate (foundation)
Knowledge Dimension: Procedural
- Isolation: sandboxes, containers, and GPU paths [Apply]
- Latency budgets per lane [Evaluate]
- Autoscaling and cold start tradeoffs [Analyze]
- Production runtime hardening [Create]
Cognitive Integration: Runtime plane depth - Rigor on SLAs.
Week 7: Observability + Eval
Bloom Level: Apply (primary), Create (secondary), Analyze (foundation)
Knowledge Dimension: Procedural & Metacognitive
Tool Mediation: OTel, LLM traces, eval harnesses
- Traces spanning tools, models, and retrieval [Apply]
- Offline and online eval strategies [Create]
- Regression suites for prompts and tools [Analyze]
- Vendor portability considerations [Evaluate]
Cognitive Integration: Observability plane - evidence for Rigor.
Week 8: Budget Enforcement
Bloom Level: Evaluate (primary), Apply (secondary), Create (foundation)
Knowledge Dimension: Metacognitive & Procedural
- Enforcing spend at control plane [Evaluate]
- Per-tenant and per-workflow budgets [Apply]
- Cost attribution and alerts [Create]
- Tradeoffs vs quality [Analyze]
Cognitive Integration: Judgment with measurable constraints.
Week 9: Progressive Delivery
Bloom Level: Create (primary), Evaluate (secondary), Apply (foundation)
Knowledge Dimension: Metacognitive & Procedural
- Safe rollout for models and tools [Create]
- Feature flags tied to policy [Apply]
- Rollback and incident drills [Evaluate]
Cognitive Integration: Production discipline - Governance in release.
Week 10: Capstone
Bloom Level: Create (primary), Evaluate (secondary), Analyze (foundation)
Knowledge Dimension: Metacognitive & Procedural
- End-to-end production agentic system [Create]
- Architecture documentation and defense [Evaluate]
- Capstone: Production agentic system with documented architecture [Create]
Cognitive Integration: Full demonstration across dimensions.
Assessment: Production agentic system with documented architecture
Track Learning Outcomes
Upon completing this track, you will be able to:
- Design agent architectures - Create effective single-agent and multi-agent system architectures
- Implement agent patterns - Apply proven patterns for agent design, communication, and coordination
- Build safety guardrails - Implement safety mechanisms, validation, and fail-safes for autonomous systems
- Orchestrate multi-agent systems - Coordinate multiple agents to work together effectively and safely
- Balance autonomy and control - Design systems with appropriate levels of autonomy and human oversight
- Implement governance in agents - Embed accountability, auditability, and explainability in agentic systems
- Handle edge cases and failures - Design robust error handling and recovery mechanisms for agent failures
- Deploy production agents - Ship agentic systems that are reliable, maintainable, and meet governance requirements
Transfer & Judgment Capabilities
This track explicitly develops two critical capabilities:
Transfer Capability
Apply learning to novel situations and contexts
- Adapt agent patterns to different problem domains and use cases
- Transfer guardrail principles across various agent architectures
- Generalize coordination patterns to diverse multi-agent scenarios
- Apply safety mechanisms to different autonomy levels and contexts
How it's developed: Through building agents for different domains, adapting patterns to new contexts, and applying guardrails to various agent types.
Judgment Capability
Make sound decisions under uncertainty
- Exercise restraint in agent autonomy decisions
- Identify and avoid unsuitable agent use cases
- Make principled tradeoffs between autonomy and control
- Balance agent capabilities with safety and governance requirements
How it's developed: Through autonomy decision scenarios, "when NOT to use agents" exercises, safety tradeoff analysis, and governed autonomy frameworks.
Track Structure
This track contains multiple levels, each with specific learning outcomes. Progress through levels to build your capability as an AI engineer.
Level 1: Foundation (Weeks 1-4)
Bloom Focus: Remember & Understand (primary), Apply (secondary)
Knowledge Dimensions: Factual & Conceptual
Capability Target: Architecture overview, control plane, policy, identity
- Architecture overview and stack framing [Remember & Understand]
- Control plane and tool registry (MCP) [Apply]
- Policy, guardrails, and safety pipelines [Apply]
- Identity and zero-trust foundations [Understand]
Learning Outcomes:
- Recalls agent architecture patterns and production stack roles
- Explains control-plane vs agent-loop responsibilities
- Applies tool registry and guardrail patterns
- Demonstrates foundational understanding of governed agentic systems
Level 2: Application (Weeks 5-7)
Bloom Focus: Apply & Analyze (primary), Create (secondary)
Knowledge Dimensions: Procedural & Conceptual
Capability Target: Durable orchestration, runtime, observability
- Durable orchestration and HITL [Create]
- Runtime plane engineering [Apply]
- Observability, evaluation, and OTel-style pipelines [Apply]
Level 3: Mastery (Weeks 8-10)
Bloom Focus: Evaluate & Create (primary), Analyze (foundation)
Knowledge Dimensions: Metacognitive & Procedural
Capability Target: Budget enforcement, progressive delivery, capstone
- Budget enforcement and cost attribution [Evaluate]
- Progressive delivery and rollout [Create]
- Production capstone and architecture defense [Create]
Assessment
- Artifact: Production agentic system with documented architecture
- Focus: Judgment, rigor, and governance dimensions
- Outcome: Ability to build governed production agentic systems
Public Sector Emphasis
For public sector participants, this track emphasizes:
- Restricted autonomy thresholds
- Mandatory human-in-the-loop
- Enhanced auditability
- Explainability requirements
Credential
Upon completion, earn the Applied AI Engineer badge - demonstrating your capability as a governed autonomy builder.
Ready to get started? Get in touch to learn more or apply.
Previous: T3: Delivery Leader | Next: T5: AI Architect