3.0: The AI Automation Framework

Key Terms

Key acronyms used in this chapter:

• AI: Artificial Intelligence
• ASPICE: Automotive SPICE (Software Process Improvement and Capability dEtermination)
• HITL: Human-in-the-Loop — a design pattern where a human is required to review, approve, or monitor AI actions at defined points
• TCL: Tool Confidence Level — ISO 26262's classification of how much qualification an AI or software tool requires (TCL1 = none, TCL3 = full)
• TI: Tool Impact — how badly a tool error could affect a safety-relevant output
• TD: Tool Detection — how likely a tool error would be caught before causing harm
• CI/CD: Continuous Integration/Continuous Deployment — automated pipelines that build, test, and deploy software on each change
• SYS: System Engineering process group
• SWE: Software Engineering process group
• SUP: Supporting processes group
• ISO 26262: Road vehicle functional safety standard
• ASIL: Automotive Safety Integrity Level (A through D)

What You'll Learn

After reading this chapter, you will be able to:

• Apply the four-level automation framework to development activities
• Select appropriate Human-in-the-Loop patterns
• Recognize AI capabilities and limitations in development contexts
• Understand tool qualification requirements for AI

Chapter Overview

This chapter presents a comprehensive framework for integrating AI into process-driven development. The framework addresses four key questions:

1. What level of automation is appropriate? (Section 03.01)
2. How do humans maintain oversight? (Section 03.02)
3. What can AI do well, and where does it struggle? (Section 03.03)
4. How do we qualify AI tools for safety contexts? (Section 03.04)

The Central Principle

Human accountability is never delegated to AI.

Regardless of automation level, humans remain:

• Accountable for decisions and outcomes
• Responsible for verifying AI outputs
• Required for judgment-intensive activities
• Essential for safety-critical decisions

AI serves as an augmentation tool, not a replacement for human judgment.

Framework Components

1. Automation Levels (03.01)

Four levels define the degree of AI involvement:

Note: Percentages are illustrative and vary by task type, team maturity, and tool capability.

2. HITL Patterns (03.02)

Six patterns ensure appropriate human oversight:

3. Capabilities & Limitations (03.03)

Understanding what AI does well—and poorly—is essential:

High Capability:

• Pattern recognition
• Text generation
• Code completion
• Consistency checking

Limited Capability:

• Novel problem solving
• Safety decisions
• Context beyond training
• Deterministic behavior

4. Tool Qualification (03.04)

For safety-critical systems, AI tools may require qualification:

Framework Application

Step 1: Identify Activity

What development activity needs AI support?

Step 2: Assess Risk

What is the impact if AI makes an error?

• High impact → L0-L1
• Medium impact → L1-L2
• Low impact → L2-L3

Step 3: Select Automation Level

Based on risk and activity characteristics, select appropriate level.

Step 4: Choose HITL Pattern

What human oversight pattern is appropriate?

Step 5: Implement with Awareness

Deploy AI with awareness of capabilities and limitations.

Step 6: Qualify if Required

If safety-critical, apply tool qualification.

Framework vs. Ad Hoc AI Use

Chapter Sections

The following sections provide detailed coverage:

• 03.01: Automation levels in detail with selection criteria
• 03.02: HITL patterns with implementation guidance
• 03.03: AI capabilities and limitations with practical implications
• 03.04: Tool qualification for safety contexts