Job Assessments: When to Grind, When to Walk

The same question comes up constantly: what part of job assessments is the most frustrating? And right alongside it, a quieter and more interesting confession — people admitting they've simply abandoned applications the moment an assessment appeared. Closed the tab. Didn't look back.

Look closely and a pattern emerges. The frustration is almost never difficulty. Nobody's angry that the puzzles are hard. The anger is about asymmetry: hours of your unpaid effort, demanded up front, by a company that hasn't yet spent a single human minute on you — feeding a scoring system nobody will explain, for an outcome nobody will report back. The test isn't the insult. The terms are. So let's do the useful thing and turn that instinct into rules. Some assessments deserve your evening. Some deserve the closed tab. The trick is telling them apart on sight.

Know what you're actually being handed

“Assessment” covers a whole family of very different animals, and your walk-or-grind decision starts with identifying the species.

The rule that sorts almost everything

Candidate effort should scale with demonstrated company effort.

Before any human at the company has spoken to you, you owe them almost nothing. After a real conversation, you can reasonably owe an evening. After several rounds with real people, a meatier take-home can be a fair trade. What's never fair is the inverted version — the marathon battery before the first hello, the unpaid weekend project as the price of a phone screen.

Run the usual horror stories through that rule and it sorts them instantly. Quick personality quiz at application? Fine — volume filter, minimal cost, do it on autopilot. Hours of timed cognitive testing before anyone has phone-screened you? That's a company outsourcing its curiosity to software and billing you for the privilege. A take-home that says “build a small working feature, should take an evening” after a good manager conversation? Legitimate, maybe even welcome — it favours people who can build over people who can chat. The same take-home before any conversation, with vague scope and no named reviewer? Walk.

And when the scope is unclear, make them commit. One message: “Happy to do this — can you confirm roughly how long it should take, who reviews it, and what the step after it is?” A serious company answers in a day, because a serious company already knows. Evasion on any of the three is your answer arriving early. Named reviewer matters most: work nobody specific has promised to read is work nobody will read.

If you decide to grind, grind smart

• Personality inventories — answer as your consistent working self, not your idealised self, and resist gaming the extremes. The instruments are dull, but they're sharp enough to catch someone alternating between saint and robot.
• Cognitive batteries — the format is most of the difficulty, so practise the format before it counts. Your first timed pattern-puzzle experience should never be the scored one.
• Take-homes — small and finished beats large and ragged, every single time. Ship the modest version, then spend your last half hour writing the notes: what you'd do next, what you traded away and why. That page of trade-offs routinely outperforms the code itself, because it's the only part that shows judgment rather than output.
• Recorded one-way videos — decide your re-record policy before you start, not mid-spiral. The trap isn't the camera; it's the retake button, which turns an evening into a perfectionism doom-loop. Cap yourself at a couple of attempts per question and keep the one where you sounded like a person. The over-polished robotic take reads worse than the human one with a stumble in it.

The opacity nobody discloses

The AI-run species deserves one more paragraph, because it's the assessment problem in its purest form: being scored by a black box on your tone, your confidence, your filler words — against a rubric nobody will show you. Not the vendors selling it as objectivity. Not the companies deploying it as efficiency. The candidate carries all of the opacity, and none of it is disclosed as a term of the deal.

And still — no call for a boycott here. AI-run assessment is going to become normal; that ship has sailed, and the people arguing about whether it should are wasting breath they could spend preparing. The current implementations are bad in a specific, exploitable way — no follow-up questions means your first answer is your only answer — and there's a full preparation guide in how to prepare for an AI interview. The one-line version: front-load every story, and practise against something that talks back and doesn't nod.

Walk loudly, not silently

Last thing. If you do close the tab — and sometimes you absolutely should — spend one extra minute doing it out loud. Withdraw formally, and say why: “I'm withdrawing my application. The assessment load requested before any conversation doesn't work for me, though I'd be glad to talk if the process changes.” Polite, specific, final.

Silent abandonment feels satisfying and teaches the company nothing — their dashboard just shows another incomplete, indistinguishable from laziness. A stated reason lands in an inbox a human reads. Enough of those and the process actually changes, because completion rates are the one metric these systems' owners genuinely fear. Plenty of people walk silently every day. If even a fraction said why on the way out, the assessments would already be shorter.