Chapter 5: How AI will affect jobs and how to stay ahead

Introduction

The chapter introduction frames software engineers’ concern about AI replacing coding work and previews the questions the chapter will address.

• Code generation and career concern: Explain why generative AI has made some software engineers and aspiring engineers worry about their careers.
  Why it matters: This concern motivates the chapter’s discussion of how AI may affect software engineering jobs.

Key terms: generative AI, software engineers

5.1 Excellence gap

The section uses the excellence gap to explain why some kinds of work are more vulnerable to AI than others.

• The excellence gap: Explain the difference between work that is merely good enough and work that aims for excellence, using the TV-writing examples.
  Why it matters: The gap gives you a way to compare different kinds of work within the same profession.
• Human work behind excellence: Describe the research, feedback, and relationship-building that can make excellent work difficult for current AI to reproduce.
  Why it matters: Looking beyond the final written output reveals work that an AI text generator may not replace.
• AI exposure and market demand: Explain how the kind of work required and customers’ willingness to pay for excellence affect a job’s exposure to AI, using the blog-writing and translation examples.
  Why it matters: Whether excellent work is possible is not enough; whether anyone demands and pays for it also matters.

Key terms: excellence gap, just fine, human touch, recipe, market value

5.2 Excellence gap in software engineering

This section contrasts routine coding with software engineering that solves business problems through collaboration and judgment, and explains why the latter may be harder for AI to replace.

• Routine versus excellent engineering: Distinguish mechanical software tasks from excellent software engineering, and explain why the author thinks the latter is better placed to resist replacement by AI.
  Why it matters: The distinction explains the section’s view of which engineering work is most exposed to AI.
• Salespeople’s control over packages: Trace how listening to salespeople changed the holiday-package algorithm project and made deployment possible.
  Why it matters: The example shows how understanding stakeholders can matter more to a project’s success than coding the initial algorithm.
• Negotiated technical tradeoffs: Describe how excellent engineers make durable and secure design choices by discussing tradeoffs with the business.
  Why it matters: Technical excellence includes choosing a solution that fits business needs, not merely maximizing features or safeguards.

Key terms: excellence gap, soft skills, business constraints, salespeople, holiday packages, future-proofing, overengineering, schemas, isolation

5.3 Recommendations

The section recommends career paths that connect technical work to business outcomes, with niche technical expertise as an alternative.

• Business Outcomes Over Code: Explain why the author recommends becoming someone who helps a business achieve its goals with technology, rather than someone who only writes code.
  Why it matters: It gives you a way to judge which technical work is worth doing.
• Cross Functional Career Choices: Describe the recommended steps toward roles and teams that combine technical and business responsibilities.
  Why it matters: These choices put you closer to deciding what should be built and why.
• Business Coupled Technical Specialties: Explain how operations research and full-stack data science illustrate technical work tied closely to business goals.
  Why it matters: They are concrete examples of careers where coding and business decisions are tightly connected.
• Niche Expertise Alternative: Describe the author's alternative for someone who prefers purely technical work and why it might help them stay ahead of AI.
  Why it matters: The section offers a possible path even for readers who do not want business-facing work.

Key terms: business goals, the right code, cross-functional teams, product management, design, operations research, data science, full-stack data scientists, niche technology, guru

5.4 Stringent validation

This section explains how the need to validate AI outputs affects which work AI can replace and whether using it saves time.

• Tolerance for imperfect outputs: Explain why companies may use unvalidated AI outputs when mistakes have little consequence, especially for useful features delivered at large scale.
  Why it matters: The consequences of an error help determine whether AI output needs a human check.
• Human validation in sensitive work: Explain why work with serious safety, legal, or commercial consequences is harder to automate away with AI, even if AI produces an initial output.
  Why it matters: Generating an answer and being able to rely on it are different things.
• Costs of checking AI work: Distinguish cases where drafting with AI and then validating it could save time from cases where checking is so burdensome that little or no productivity gain remains; describe the possible pressure to pay validators less.
  Why it matters: The time and cost of checking an AI result belong in any estimate of its productivity benefit.

Key terms: stringent validation, nice to have, hallucinate, human expert, involved in the loop, productivity gains, validation, AI-generated text

5.5 Validation in software engineering

This section explains why AI-generated code still needs thorough validation and why software engineers remain important even if their work shifts toward reviewing it.

• Software validation stakes and practices: Explain why software is validated thoroughly and describe the validation practices named in the section.
  Why it matters: The consequences of a small bug explain why code needs checks before people rely on it.
• Limits of AI-generated code: Distinguish code that compiles from code that works correctly, using the traffic-map and weather-map experiments to explain why AI output needs debugging.
  Why it matters: Working-looking code is not proof that an AI solution meets the task or handles real data correctly.
• Engineer expertise and job roles: Explain why the author thinks software engineers remain necessary with AI-generated code and describe the possible change in how companies value their work.
  Why it matters: AI may change engineers’ work and bargaining position without removing the need for their expertise.

Key terms: unit-test, code review, automated tests, quality assurance (QA), syntactically correct, runtime, API, detailed prompts, debug, validate

5.6 Recommendations

The section recommends validation-centered software practices and systematic AI product benchmarking as ways to deliver reliable products and remain valuable when AI can write code.

• Robust Software Delivery: Explain why delivering robust software is a stronger professional focus than writing new feature code alone, and describe how validation-driven teams support it.
  Why it matters: It gives you a way to contribute beyond the code AI can generate.
• Test-Driven Development Practices: Describe how the section recommends using tests around small units of code, and explain the balance it calls for.
  Why it matters: Tests left in the codebase can catch problems again later.
• Trunk-Based Development Safeguards: Distinguish trunk-based development from the separate-branch workflow described here, and explain how the section proposes protecting users while changes are validated.
  Why it matters: It shows how continuous delivery still requires deliberate validation safeguards.
• AI Product Validation: Explain why AI-based products need systematic validation and describe how benchmarks and smaller subtasks can help improve them.
  Why it matters: Measured performance makes AI product problems easier to locate and improve than a single prompt and hope for the best.

Key terms: robust software, validation driven, DevOps, test-driven development, TDD, edge cases, test case, trunk-based development, main code branch, feature flag

5.7 Tight control

This section explains why generative AI is more useful when precise control of its output is unnecessary, and how that difference affects which jobs are more exposed to AI.

• Output control and prompt effort: Explain why generative AI can save time when an acceptable result matters more than precise details, but may save little time when the output must match exact requirements. Contrast the quick-ad graphics with the graphic designer’s difficult image-generation task.
  Why it matters: How much control a task requires helps you judge whether AI will actually make it faster.
• Job exposure to AI: Explain why jobs whose clients care little about precise output details are more exposed to AI than jobs with demanding, specific requirements. Use the quick-ad designer and high-end visual effects artist as examples.
  Why it matters: The client’s need for precise results can matter more to job exposure than the job title alone.

Key terms: tight control, prompt, good bang for your prompt, precise requirements, replaced by AI, resist AI

5.8 Control in software

The section contrasts a loosely specified coding task where AI was useful with customized software work that still requires precise technical instructions.

• Flexible coding tasks: Explain why ChatGPT was useful for the line-chart color palette despite the brief prompt and unattractive colors.
  Why it matters: It identifies the kind of coding task where a quick AI-generated result can be enough.
• Precision in customized software: Explain why asking AI to produce highly customized software can become much like programming, and why engineers still need precise technical language.
  Why it matters: It explains why AI assistance does not eliminate the need to specify what complex software must do.

Key terms: color palette, prompt, exact output, requirements, AI prompt, precise, step-by-step technical language

5.9 Recommendations

The section recommends choosing software work where precise requirements matter over disposable prototype work that clients may increasingly do without engineers.

• Choosing detail-sensitive software work: Explain why the author recommends avoiding jobs building quick, disposable prototypes and seeking demanding projects for clients who care about small details.
  Why it matters: It gives you a way to distinguish software jobs clients may do themselves from work where exacting requirements matter.

Key terms: toy products, quick prototypes, no-code app builders, demanding projects, picky clients

5.10 A new opportunity: Making the web more human

The section argues that demand for genuine human content may grow as AI-generated content spreads, creating opportunities to rank and filter content by authenticity.

• Demand for authentic content: Explain why people may seek human opinions rather than ordinary search results, and how a flood of AI-written content could create jobs focused on content authenticity.
  Why it matters: It connects a frustration people already have with online content to a possible new kind of work created by generative AI.

Key terms: genuine, human content, AI-written content, authenticity, information retrieval

5.11 Philosophical detour: Automation and mass unemployment

The section argues that automation can raise long-run economic output while imposing painful costs on displaced workers, then applies that distinction cautiously to generative AI.

• Automation and economic output: Explain how the section argues that efficient automation can make an economy richer: producing essentials with fewer work hours frees resources for other products and services, increasing output rather than causing prolonged mass unemployment.
  Why it matters: It explains why eliminating particular jobs does not necessarily mean the economy has less work or less output overall.
• Costs for displaced workers: Distinguish the section’s claim about long-run economic gains from its account of workers who lose jobs and may have to seek lower-paid work because their skills are no longer needed.
  Why it matters: It keeps an economy-wide argument from dismissing the real harm experienced by displaced workers.
• Qualified outlook for generative AI: Describe why the author expects generative AI to support some economic growth without treating it as an unqualified productivity breakthrough: it helps with certain tasks, but hallucinations, validation needs, and resource use limit its efficiency.
  Why it matters: It applies the section’s economic argument to AI without overlooking costs or overstating its benefits.

Key terms: automation, resources, real GDP, obsolete skills, transition, generative AI, hallucinates, validated

5.12 Summary