Staying valuable as AI commoditises skills and knowledge work

Executive overview

AI is turning writing, coding, and design into push-button tasks, collapsing the value of isolated skills. The people who stay valuable are those who maintain a gap between what AI can do alone and what they can direct AI to do.

The gap isn't fixed — it moves upward as models improve. Staying ahead means mastering techniques that let you get more complex outputs faster, and developing the judgment to deploy AI resources where they matter most.

The durable edge is not deeper expertise in any one domain — it's the breadth and taste to direct AI better than everyone else.

Techniques that separate fast builders from slow ones

• Use a guiding technique: a small reasoning model (e.g. o4-mini) interviews you to produce a PRD; a stronger model (e.g. o3) converts it into a blueprint and task list.
• Know AI-centric IDEs (Cursor, Windsurf) well enough to set accurate project and global rules.
• When a model isn't inside your IDE, use tools like YEEK to flatten your codebase into a text file and feed it to an external model for debugging.
• Use the root-cause technique: ask AI to assess errors without changing anything first, then self-reflect before acting.
• Apply test-driven development — have AI write unit tests before writing functionality, then loop until tests pass.
• Start fresh conversations when the context window degrades; pass forward only the relevant prior context.

Taste: matching models to use cases

• Hundreds of models exist; in practice, 6–8 matter for most work.
• Model selection is a skill: 4.5 Pro suits writing, 2.5 Pro suits planning and gnarly debugging, 2.5 Flash suits cheap reasoning and data extraction.
• Good taste comes from heavy use across scenarios plus reading how others apply the same tools.
• Wrong model for the task wastes time and produces weaker outputs — matching is a real lever.

Breadth over depth

• Going deep is increasingly outsourced to AI; going wide is the human advantage.
• Breadth lets you know what questions to ask — AI can't yet proactively guide you, so the quality of your prompt depends on knowing what's possible.
• Understanding available tools (deployment, databases, data pipelines, scraping) lets you direct AI rather than wait for it to suggest directions.
• Tomorrow's edge is spotting opportunities: knowing where to deploy intellectual capital before competitors do, the same way investors decide where to deploy financial capital.