"Taste in the Age of AI: Selection as a Core Skill"

According to a report from Huxiu on “AIGC: From Zero to One,” AI-driven video generation technologies are rapidly advancing, producing a growing amount of content that, while seemingly flawless on a surface level, often lacks the qualities that make one want to share it after viewing. This phenomenon has reignited interest in the concept of “taste” or “sense.”

The Resurgence of the Debate on Taste

In early 2026, Paul Graham, founder of Y Combinator, wrote, “When everyone can make anything, the real difference lies in deciding what to make.” Two days later, Greg Brockman, former president of OpenAI, stated, “Taste is the new core skill.” Adding to the conversation, Cloudflare’s CTO Dane Knecht identified “sense” as a key differentiating factor in engineering by 2026.

Huxiu’s report analyzes these statements, noting that while they all point in the same direction, they emphasize different aspects. Graham focuses on the power of choice. In an era of AI-driven production, those who determine what to create first hold the reins of direction. Brockman highlights changes in the nature of work—prioritizing selection, sequencing, and setting priorities over mere execution. Knecht’s engineering “sense,” by contrast, boils down to concrete decision-making: whether to add layers to an architecture, strip down features, or allocate resources to particular user experiences.

At the same time, skepticism about this movement exists. Nan Yu, Product Lead at Linear, argues that not everyone necessarily possesses superior taste compared to AI. Moreover, Matt Schumer observes that new models can present choices that resemble human judgment, rather than merely following instructions. These critiques suggest that the term “taste” might be too ambiguous, potentially serving as a convenient buzzword to feign expertise.

The Core Definition of Taste

Huxiu’s article distills the discussions about taste into three distinct capabilities: “choosing what’s worth doing,” “doing things correctly,” and “knowing when to pull back.” While interconnected, these aspects cannot be reduced to a simple appreciation for aesthetics.

The article defines taste as “the ability to make selective decisions from a vast array of possible options based on specific circumstances and to take responsibility for those choices.”

A critical element of this definition is the notion of “possibilities.” AI can generate vast amounts of content by combining existing patterns, much of which technically meets the mark. However, discerning which content is worth pursuing, which to abandon from the outset, and which to prioritize requires a judgment that goes beyond mere generative ability.

Personal Preferences vs. Judicious Decision-Making

Understanding taste as mere aesthetic preference is not entirely incorrect. Some people prefer minimalism, while others enjoy excessive ornamentation. However, the “taste” discussed in the age of AI fundamentally differs. It refers to the ability to sift through a multitude of seemingly adequate options and identify what best suits the current context, what is merely a familiar pattern, what should be discarded, and what holds potential for further development.

This judgment transcends aesthetic considerations, manifesting in fields like product design, engineering, and writing. A product manager might cut a seemingly cool feature that complicates the main user flow. A director might decide to exclude technically flawless footage. An engineer might refuse to add unnecessary architectural layers. These decisions all exemplify the exercise of taste.

The article notes that the value of taste lies not in its resistance to imitation but in its ability to define what others should imitate. In the age of AI, this concept carries particular weight, as AI models extract patterns from existing creations and recombine them. Before AI-generated options advance further, someone must decide which are worth pursuing, which are superfluous, and which should never have been created in the first place. Taste governs this gateway.

The Proliferation of Passable Content and

Resistance to Mediocrity

One of AI’s transformative effects is the sheer volume of “adequate” content it produces. In the past, creating a full-length video required scripting, shooting, editing, and narrating. Now, this process has been compressed into prompts and a few revisions. As a result, content has lost its scarcity, and attention spans are consumed more quickly than ever.

The article describes the current situation as “not a flood of low-quality content, but an overwhelming abundance of passable content.” The rhythm is flawless, the information complete, the emotions well-placed—yet the reason these creations are swiped past is that every element feels calculated from pre-validated logic. “Adequate” has become a remarkably low bar.

Taste’s role in this context is to prevent people from being dragged down by mediocrity. A video may be smooth but lack genuine insight. A proposal might be thorough but miss the core issue. A product might be functional but unnecessary.

This ability is not glamorous. It often means removing more than half of the content, admitting a first draft is unworthy of revision, or saying “let’s not do it” when everyone else is saying “we can.” In this sense, taste resembles editing skills—good editing doesn’t embellish an author’s writing but identifies which parts obstruct the article’s message.

The Conditions That Make Taste Difficult for

AI to Mimic

While AI can learn to identify existing styles perfectly, it struggles when encountering new and unfamiliar elements that lack established evaluation criteria. In such cases, a willingness to make decisions despite the risk of being misunderstood is essential.

The article argues that truly valuable taste often involves elements that don’t align with current trends and mixes personal experience with judgment, making it difficult for AI to distill into a replicable style list. This insight is crucial for understanding cooperation with AI.

Huxiu suggests that mature taste is both discernible and accumulative. When decisions impact others or resources, the reasons behind those choices must be clear. Moreover, mature taste emerges from metrics accumulated through extensive practice. Only after experiencing numerous errors and redundancies can one establish accurate criteria for judgment.

Practicing Taste in Collaboration with AI

When collaborating with AI, humans should focus less on the initial generation and more on the stages of selection and judgment. Specifically, they must set boundaries for selection, constantly ask “why choose this,” and intentionally forego viable alternatives.

Ultimately, taste is not a fixed standard but a constantly calibrated process, rooted in the willingness to take responsibility for being “wrong.” This perspective aligns with the broader discussion about monitoring AI agents’ decisions. As AI evolves from a tool into a decision-making partner, the human role is shifting from “generation” to “selection.”

Huxiu concludes that taste is not a skill that can be replaced by AI but rather a capacity to direct the output of AI. Amid the flood of AI-generated creations, humans must protect the value of choice—and at the core of this responsibility lies the concept of taste.

Editorial Opinion

The mass production of passable content by AI is expected to accelerate within the next 3 to 6 months. In this environment, the ability to discern and select outputs may prove more critical for product teams and engineering organizations than mere proficiency in using AI. With rapid prototype generation becoming easier, early missteps in direction could lead to substantial costs. In our view, redefining taste not as “individual aesthetic preference” but as “organizational decision-making criteria” is an essential discussion.

From a long-term perspective, over the next 1 to 3 years, AI might begin to offer choices closer to human judgment. Some models are already reported to be able to provide suggestions beyond mere execution. However, AI judges new entities using existing evaluation frameworks, while human taste includes “out-of-step decisions” and “the willingness to support risky innovations”—qualities not easily replicated by AI. The editorial team is closely monitoring how the division of roles between humans and AI evolves, potentially shifting from “generation vs. selection” to “existing vs. new.”

Finally, we pose the following key questions for further reflection:

References

• ” 当我们谈论品味的时候，到底在谈论什么 ”, by AIGC从0到1 — 虎嗅网, 2026-07-13T19:04:32.000Z (ARR)
• Source URL: https://www.huxiu.com/article/4874983.html?f=rss