The Half-Life of Engineering Skills

The most dangerous career move right now? Getting exceptionally good at something AI is quietly learning faster than you.

I was talking to Vijay Bhaskar and Rakesh Pandey, and Vijay brought up a curious phrase (and an extremely useful concept!):

"Half-life of skills!"

Like physics… except instead of uranium, it's your resume decaying.

Some skills last a decade. Some last a few years. Some expire midway through your Udemy course. Take prompt engineering.

• Last year: "critical skill"
• This year: baked into tools
• Next year: we will explain it to interns right after "what was a fax machine?"

And this isn't just vibes. Research from IBM, LinkedIn, and World Economic Forum all point the same way:

• Skills used to last ~10+ years
• Now closer to ~4–5 years
• In AI-heavy work: ~2–3 years (and dropping faster than my attention span in long meetings)

We're not learning skills anymore. We're renting them.

So what actually survives?

Long half-life:

• Deep domain understanding
• Customer intuition
• System design thinking
• Clear problem definition
• Judgment (famously difficult to automate… mostly because no one can define it)

Short half-life:

• Tool expertise
• Prompt tricks
• Workflow hacks
• Anything that can be turned into a button

Here's the uncomfortable shift: AI is compressing execution skills …and quietly raising the price of thinking skills. Most organizations are still optimizing for faster execution. The real leverage is shifting to: better intent + better evaluation.