Skip to content

6.0 Introduction to Part 6: Artificial Intelligence

Artificial intelligence has gone from research novelty to a core capability that large organizations are now expected to deploy responsibly and at scale. This part walks through the whole lifecycle: deciding where AI is worth applying, engineering machine learning systems that survive contact with production, building applications on generative models, using AI to help with software development itself, meeting your obligations for fairness and accountability, and running the specialized infrastructure that all of this needs. One idea ties every chapter together: the model is rarely the hard part. The hard part is the engineering, governance, and operational discipline that surround it.

For large teams, scale and inertia raise the stakes. A poorly framed initiative can burn budgets, distract skilled engineers, and erode trust when it fails in public. A well-chosen one can automate drudgery and surface insight from data you could never reach before. Across hundreds of developers and dozens of models, small habits and small gaps in oversight add up to organizational outcomes. That is why shared platforms, defaults, and norms matter more than any single clever solution.

Enterprise and government contexts add sharper constraints, and they run through every chapter. Public bodies must justify spending, guarantee transparency, avoid unlawful discrimination, and stay accountable to regulators and the public. Procurement rules may forbid sole-source lock-in and demand explainability and exit options. Emerging law such as the EU AI Act (the European Union's risk-tiered regulation of AI systems) and standards such as the NIST AI Risk Management Framework and ISO/IEC 42001, a certifiable AI-management-system standard, impose requirements graded by risk. In these settings, compliance, auditability, reproducibility, and data sovereignty are first-class requirements, not afterthoughts.

Chapters in this part

  • 6.1 AI strategy and readiness: The discipline of deciding where to apply AI, where to avoid it, and what data, talent, and governance foundations must exist before the first model reaches production.
  • 6.2 Machine learning engineering (MLOps): Bringing software rigor to the three-part reality of code plus data plus models, so ML systems are reproducible, observable, and maintainable rather than one-off demos.
  • 6.3 Generative AI and LLM applications: Engineering dependable applications around probabilistic language models through context management, retrieval-augmented grounding, guardrails (runtime checks that constrain model inputs and outputs), and rigorous evaluation.
  • 6.4 AI-assisted software development: Using AI coding assistants to raise throughput without lowering quality, through clear norms, good defaults, strong verification, and attention to security and licensing.
  • 6.5 Responsible and trustworthy AI: Building systems that are fair, transparent, accountable, safe, and privacy-respecting, and being able to demonstrate all of this to those affected and to regulators.
  • 6.6 AI infrastructure and operations: Provisioning and running the scarce accelerators, vector storage (systems that index data as numeric vectors for similarity search), and serving layers AI workloads demand, cost-effectively, reliably, and observably.

How these chapters interrelate

Together these chapters form a lifecycle that runs from intent to operation. Chapter 6.1 sets strategy and readiness, framing which problems justify AI at all and what foundations you need first. Chapter 6.2 supplies the engineering backbone for classical machine learning, while chapter 6.3 covers the distinct patterns of generative models, where behavior is shaped by prompts and retrieved context rather than a training step. Chapter 6.4 turns AI on the craft of software development itself, and chapter 6.6 provides the compute, storage, and serving substrate that every other chapter depends on. Chapter 6.5 is deliberately not the last word but a constant companion: responsible-AI obligations shape strategy, constrain model choice, gate deployment, and inform infrastructure decisions throughout.

The throughline runs from strategy to MLOps to generative AI to AI-assisted development to responsible AI to infrastructure, but think of it as a web, not a straight line. MLOps and infrastructure share concerns around reproducibility, cost, and serving, so chapter 6.2 and chapter 6.6 are best read together. Responsible AI ties back to strategy in chapter 6.1 and forward to the broader treatment of ethics and public interest in chapter 10.5, and it leans on the privacy and data-protection practices in chapter 4.5. The wider engineering foundations of this guidebook underpin the whole part: continuous delivery in chapter 8.1, observability and monitoring in chapter 9.2, testing strategy in chapter 2.4, and code review in chapter 2.5 are all prerequisites for doing AI well. AI does not replace sound software engineering. It raises the stakes on getting it right.