The short version
- Taste is the noticing, and the naming comes second. The naming is teachable. I have never found a timeline for the noticing.
- Taste that survives a measurement is taste. A composited product screen was called fake before anything was counted, and the count came back 0 usable frames against 3 of 3.
- Test it on a real artifact with no brief attached. Use one that passes every mechanical check and still fails, so a candidate reciting a checklist has nothing to recite.
- Do not hire taste to do a script's job. An audit of 200 essays here found 2,801 em-dashes against a house rule of zero.
Taste is noticing that something is wrong before you can explain why, then working out the explanation. Craft improves on a schedule you can plan around. This does not, which is why I weight it above the parts of a job I know how to teach. The test is whether the noticing survives being measured.
What does taste look like in the work?
Somebody looks at a frame and says it reads as a sticker. On one account we needed a product screen visible on a phone inside a vehicle. The composited version had a warped screen sitting proud of the bezel, no glow on the hand holding it, no grain shared with the plate underneath.
Nobody itemises that list in the moment. They say it looks fake. The itemising happens afterwards, and it is what makes the judgment transferable, because a list can go into a brief and a feeling cannot.
What separates the two, and what each gap costs to close:
| Craft | Taste | |
|---|---|---|
| What it is | Producing the thing to a stated standard | Noticing the standard is wrong or missing |
| How it shows in a work sample | The sample meets the brief | The candidate says what is wrong with the brief |
| What its absence produces | Work that misses the bar | Work that clears every check and has no point |
| Can a script cover the gap | Often, for anything with a right answer | No. A count cannot tell you a page is empty |
| How long the gap takes to close | A timeline you can put in a plan | Not a timeline I know how to write |
How do you tell taste from opinion?
Taste survives a measurement. The sticker call was a judgment made before anything was counted, and then it was counted: 0 usable frames from the composited method against 3 of 3 generated in one pass.
An opinion that will not name a test is a preference. Somebody who calls a page thin and then names what to count on it, rendered length and how many of its numbers are first-party, has turned the feeling into a check anybody can run.
How do you test for it?
Hand over a real artifact with no brief and ask what is wrong with it. The best one passes every mechanical check and still fails. A batch of twenty posts written for this site came back clean in structure, around 595 words against a bar of 1,000 to 1,400, with no first-party data on any of them.
The answers separate on whether the candidate can price the fix, not on whether the fault gets spotted. Spotting is common. Naming what correction costs, and deciding some of it is not worth correcting, is the part being hired.
Where is taste not the constraint?
The mechanical half. An audit of 200 essays on this site found 2,801 em-dashes against a house rule of zero. Nobody decided to ignore it. It was written down and restated in briefs, with nothing behind it that could fail a build.
Anything with a right answer belongs in a validator, and hiring judgment for that job wastes the person while leaving the problem in place. Choosing the quality test is itself a taste decision, and reading a draft for whether it sounds like AI was the wrong one.
Nobody says the screen has no glow on the fingers holding it. They say it looks fake, and somebody works out why afterwards.
What does taste look like when it costs money?
Knowing when to stop. Forcing a right-hand-drive interior out of an image model took 5 attempts, because the prior for a car interior is left-hand drive. The fix was changing the composition to keep the console out of frame.
The person I want changes the shot on the second attempt rather than the fifth, because the decision is arithmetic about how many calls a frame is worth.
The two engagements underneath this one are why a composited screen reads as a sticker and the half of quality a script should own.
Frequently asked questions
What does hiring for taste mean?
Hiring for the ability to notice that something is wrong before anybody can explain why, and to name the reason afterwards. It shows up as the person who calls a page clean and empty when every check on it has passed.
How do you test for taste in an interview?
Hand over a real artifact with no brief and ask what is wrong with it. Use one that passes every mechanical check and still fails, so a candidate reciting a checklist has nothing to recite.
Can taste be taught?
The naming can. Somebody who says an image looks fake can be taught to itemise the missing glow, the mismatched colour temperature and the sharp edge sitting on grain. The noticing has no timeline I know of.
Is taste more important than experience?
For roles that produce judgment rather than volume, yes. Experience predicts how fast somebody reaches a stated bar. Taste predicts whether they notice the bar is in the wrong place.
What is the difference between taste and opinion?
Taste survives being measured. Calling a composited product screen fake was a judgment first, then it was counted: 0 usable frames from that method against 3 of 3 generated in one pass.
How do you hire for taste without hiring people who think like you?
Score the answer rather than the agreement. The signal is a candidate naming a fault you had missed in your own artifact. If every strong candidate agrees with you, the test is measuring similarity.
Sources
- Chua Network delivery data across 8 client accounts (internal fact bank)
- Chua Network engagement records, anonymized (internal experience bank)
