A Degree Is Not Job Competence

There has to be a marker for how much education the candidate got. Model tiers are that marker.

Opus, Sonnet, Haiku — or any other vendor's lineup — differ in general caliber: how deep the reasoning goes, how broad the knowledge spreads, and the price tag that goes with it. This is exactly what a degree does on a resume: it quickly tells the employer roughly what league the candidate plays in.

The marker is earned by examination. The candidate's tests during school and the standardized exams at graduation correspond to the model's general scoreboards — the benchmarks and leaderboards. The diploma is the result; the leaderboard ranking is the process. Both belong to pre-employment evaluation, the kind that happens while still in school. It measures general caliber, and it happens before any particular job exists. This exam sits between birth and graduation, not after hiring. After hiring comes another exam, one that tests job competence — that is what evals are for.

A degree is an input, not a conclusion

But a degree has never translated directly into job competence.

A PhD is not necessarily a better fit than a master's for a given role. For a simple classification job demanding fast responses at a thousand calls a day, hiring a PhD is using a sledgehammer to crack a nut — slow and expensive. For an architecture role demanding long chains of reasoning where one wrong step poisons every step after, sending an intern is trading cost savings for incidents. A degree's level speaks to general ability; a role demands a match — how deep must the reasoning go, how much error can be tolerated, how much latency and cost can be paid.

Put this back into model selection and it becomes clear: picking a model is not "stronger is better" but "matched to the role." Assigning the most expensive model to a once-a-day format conversion is waste. Putting the cheapest one in charge of core decisions is a gamble. A strong model in the wrong place burns money and is not necessarily better; a weak model in the wrong place saves money that will not cover the damage.

Matching requires a job first

The problem is that a match needs something to match against. You have to know what the work requires before you can judge which tier of degree fits.

And what the work requires is exactly what most people have not figured out when they pick a model. Grab the strongest model first, then work out what to have it do — that gets the order backwards. A degree is one input to the role match, not the match's conclusion; without a job definition, even the highest degree has nowhere to land.

So the next step is not more model comparison. It is writing the job down: write the JD first, then pick the model.