The engineering skills that matter more in an AI world

AI is brilliant at generating code quickly. Give it requirements for a notification service, and you'll have working code within an hour. The structure looks good, it compiles without errors, and on the surface everything looks fine.

This reveals something important: if a task can be fully specified and follows established patterns, AI can handle it. The real value lies in what AI can't replicate.

Skill 1: Knowing good code from working code

Recently I used AI tooling to build out a feature and found a single method that was 300 lines long, handling an API call, emails, text messages, database updates and error logging in one block. Technically it worked, but maintaining this would be a nightmare.

Spotting this requires experience. When I write code, I'm constantly making decisions that don't appear in requirements documents. Can this handle increased traffic? What happens when a third party is unavailable? How easy will this be to fix in five years?

Evaluating code quality beyond mere functionality is becoming one of the most valuable engineering skills. AI meets today's requirements but has no thought for tomorrow.

Skill 2: Knowing what to fix (and what to bin)

I spend time refactoring AI-generated code until I'm comfortable releasing it. The tool saves hours on initial setup, but I invest time making it correct. Still a net positive.

What's significant is that refactoring AI code effectively is itself a skill. Code review has always been important, but in an AI-assisted world, it's the critical checkpoint. The ability to quickly identify issues and guide improvements increasingly separates effective engineers from the rest.

Skill 3: Understanding context

The codebases I work on have conventions evolved over years, driven by how systems work together, requirements, and past mistakes. AI has no understanding of this context, even with full codebase access. It gives generic solutions when we need tailored ones.

This contextual understanding (knowing why systems were built certain ways, what constraints drove decisions, how changes ripple through an organisation) is perhaps the most irreplaceable engineering skill.

Skill 4: Knowing when to use AI

I've developed a mental map of where AI excels and where it struggles. This pattern recognition matters when deciding whether to leverage AI or rely on human expertise.

AI consistently delivers on boilerplate, data transformations, and test generation. It struggles with complex state management, performance optimisation, security considerations, and anything requiring deep contextual understanding.

The pattern is clear. AI excels at localised, well-defined problems. When a task requires understanding how multiple systems interact or why specific architectural decisions were made, AI becomes a starting point, not a solution.

Knowing when to use which approach is an increasingly important engineering skill.

Skill 5: Choosing the right approach

I use AI as a thought partner. When stuck, I ask it to suggest approaches, then pick the one that suits our needs best. This selection process is distinctly human: evaluating options, understanding trade-offs and choosing the right one for your context.

These AI tools are like an extremely knowledgeable but inexperienced colleague who works fast, but you wouldn't trust with important decisions. The strategic thinking and architectural judgement calls remain firmly with experienced engineers.

What This Means for Engineers

My job isn't just to write code that works now. It's to write code that works when requirements change, that colleagues can understand, and that's easy to maintain in years to come. That takes judgement, experience, and contextual understanding that AI doesn't have.

Rather than replacing engineers, AI is clarifying what makes us valuable. The skills that matter most are distinctly human: recognising quality beyond functionality, effective code review, contextual understanding, knowing when to use AI, and strategic problem-solving.

Here's the uncomfortable truth. Engineers who refuse to adapt will find themselves at a serious disadvantage. Not because AI will replace them directly, but because engineers who've mastered working with AI will be significantly more productive. The gap between those who embrace these tools and those who resist them is widening rapidly.

I'm not worried about being replaced. I'm excited about what I can build with these tools whilst being clear about what makes me irreplaceable. The future belongs to engineers who understand both the power and limitations of AI.