Created on 2025-10-11 10:34
Published on ---
Stop treating taste as magic. Start designing it
Last spring I walked a gallery with five colleagues. Marcus, our unofficial oracle, trailed behind with the curator. We stopped at a piece: a shopping cart filled with concrete, hung by fishing line, a single red sneaker on the handle.
We roasted it for ninety seconds. Trying too hard. Too obvious. Then Marcus arrived. He stared, nodded, and said: “The banality of consumer infrastructure made unbearable. It is hard to look at.”
Faces recalibrated in real time. The art did not change. The context did. Our taste followed. You have probably done this too.
Taste as compressed context
People say: as AI takes execution, humans keep taste. That sounds reassuring. It is also vague. If we cannot define taste, we cannot protect it or improve it.
Here is a better frame: taste is compressed context. It is the residue of exposures, incentives, constraints, and goals, compressed into expectations about what “good” looks like. If this is true, taste is learnable and designable. Our edge will not be that we “have taste” and machines do not. It will be that we choose the contexts worth learning from.
What this predicts
* Change the label, change the judgment.
* Stable contexts create stable taste.
* Shared contexts create convergent taste.
* Training with feedback creates finer discrimination.
Biology sets guardrails. Context writes most of the code. If you want better taste, design better contexts.
Three fast tests (do these this month)
1. Blind vs branded Show the same work twice. First, stripped of labels and backstory. Then, with full branding and price. Watch for: ratings that flip when the context arrives. If the expensive option suddenly looks better when you reveal the price, you just saw taste-as-context. Move: judge blind first; add context second.
2. Story swap Take one piece of work. Write two origin stories: one craft, one speed; one mission, one commercial. Show each to different groups. Watch for: the same work earning different adjectives based only on narrative. Move: decide which story you want to import into decisions.
3. Constraint sprint For two weeks add one constraint to every decision: under 500 dollars, zero waste, ships in 48 hours. Watch for: your definition of “good” bending toward the constraint. Move: use constraints to bend taste on purpose.
Why this matters now
Agentic AI is moving from demos to daily tools. Systems that plan, call functions, and execute workflows require explicit objectives, criteria, and feedback. They do not operate well in contradiction or ambiguity. The clearer your contexts, the better they perform. The fuzzier your contexts, the noisier the outcomes.
Working with AI
If taste is compressed context, then systems trained on vast, structured contexts can approximate it. That is not a threat. It is a division of labor. AI surfaces patterns from massive context. Humans set goals, constraints, and narratives. When you treat taste as designed inputs rather than magic, collaboration improves: prompts become specifications; reviews become tests; “good” becomes measurable enough to iterate.
Skills to build now that taste is designable
* Curate contexts worth learning from.
* Name the constraints that matter.
* Accept accountability for the inputs you choose.
Start this week
* Audit inputs: list your top twenty sources and rooms. Prune and upgrade.
* Run a constraint sprint: pick one filter for fourteen days.
* Explain each pick: one sentence per “yes.” The language reveals hidden context.
* Study extremes: your best and your worst side by side. Annotate what moves you.
* Rotate rooms: spend time with a different community. Notice how “obvious good taste” travels.
Bottom line
The future of work will not reward people who “have taste.” It will reward teams that design the contexts that produce it and take responsibility for those choices. That is not only compatible with AI. It is the lever that makes the collaboration smarter.