The 80 vs 20

I’ve complained about the interview process for quite some time.  It’s an open secret in our industry that the way we hire software engineers produces absurd, wildly inaccurate outcomes.  A quick browse on linkedin pretty much will confirm that.

The 80/20 Reality of Engineering
The fatal flaw of the traditional technical interview is that it tests for the wrong skills: raw fact regurgitation and greenfield code generation.

Think about the actual day-to-day life of a software engineer. How much of your day is spent staring at a blank IDE, typing out brand-new algorithms? Very little.

In reality, an engineer’s time follows a pattern of 80/20:

80% of the day is spent reading existing code, debugging legacy systems, tracing data flows, and analyzing complex architectures.

20% of the day is pure writing.

Yet, our interview processes test almost exclusively for that 20%. We treat software engineering like a high-speed typing and memorization contest, rather than an exercise in deep system comprehension.

AI may be the Game Changer

Now, throw AI into the mix. As we all know, AI writes the code. The nature of "writing" software has fundamentally shifted.

Instead of wrestling with syntax, engineers now spend more time writing a spec (especially if you are leaning into loop engineering or spec-driven development), reviewing the structural plan the AI created, and rigorously auditing the AI’s output.

The premium skill is is knowing what to build, how it fits into the broader system, and why a piece of generated code might fail in production.

So, does it even make sense to do pure coding interviews any more?  Pure coding challenges shouldn't be the gatekeeper at the front of the hiring funnel. If we use them at all, they should be reserved as a minor tie-breaker to differentiate between two or three viable final candidates. The upfront part of the hiring process needs to adapt to reality. It should focus on how a candidate reads code, how they critique a system design, and how they manage the AI loop.

I hope that is the case. Because if AI finally forces the tech industry to stop testing for the 20% and start testing for what engineers actually do all day, the interview process might finally be fixed.