Winning in the AI Product Era: The Four Discriminative Skills That Matter

1. From Desktop Publishing to AI‑Powered Creation

In 1984 Apple democratized desktop publishing, letting anyone design print material, but Steve Jobs later noted that 90% of the output was ugly because tools were democratized while taste was not. Forty years later AI agents, low‑code platforms, and AI IDEs have similarly lowered the cost of building an MVP, but the core problem remains: creating is easy, making something good is hard.

Industry writers (The New Yorker, Paul Graham, Axios) now treat “taste” as the new competitive dimension in the AI era. When technology is no longer a barrier, taste becomes the gate‑keeper.

2. What Is "Taste"?

Taste is not merely aesthetic appeal; it is the ability to judge whether a product should exist. Paul Graham’s Taste for Makers links good design to simplicity, timelessness, solving the right problem, and meticulous detail. For AI product managers, taste manifests in four discriminative abilities:

1) Distinguish Stated Needs from Real Needs

Example: Users ask for an “AI portrait” but truly want a better view of themselves. Many AI‑portrait apps focus on adding more style templates, yet the real pain point is that the generated photo doesn’t look like the user.

2) Distinguish Features from Value

Stacking ten mediocre AI capabilities does not equal value. Users often need a single core workflow; extra features add cognitive noise.

3) Distinguish Flashy from Truly Useful

Notion’s quiet “Ask AI” button is less flashy than a conversational UI but proves far more usable. AI can generate dazzling 3D transitions, but if they slow tasks down they are merely flash.

4) Distinguish Personal Preference from What Users Will Pay For

A minimalist design may please a designer, but older users in lower‑tier markets may prefer dense information. The right judgment aligns with the target audience, not the creator’s taste.

3. How to Cultivate Taste

Taste is not innate; it can be trained through deliberate practice.

Step 1: Massive Input, Not Just Watching, But Dissecting

Review 100 similar products and ask: What does the first screen teach? What drives users to stay, pay, or feel cheap? Reverse‑engineer the decision‑making behind each element.

Step 2: Force Deletion

Adding is easy; deleting requires admitting a idea is wrong. AI makes generation cheap, but the “AI‑generation cost fallacy” means we still resist removal. Ask: If this feature were removed, would users notice?

Step 3: Build a Structured Sample Library

Collect UI patterns (hero sections, payment pages, empty states, error messages, onboarding flows) and tag them by scenario, emotion (trust, surprise, safety, premium), and user stage (new, active, churn‑risk). This library becomes a reference for AI prompts.

Step 4: Calibrate with Real User Feedback

Behavioral data (where users linger, drop off, pay, share) is more honest than surveys. Use it to adjust the discriminative criteria.

4. Why “Volume‑Based Betting” Fails

Some teams launch dozens of AI products hoping one will explode. This confuses “trial‑and‑error” (hypothesis‑driven learning) with “mass‑betting” (random attempts). AI lowers the creation barrier but not the distribution, ops, compliance, or taste barriers. Ten unused products are worse than one product with 100 loyal users.

Mass‑betting also amplifies homogeneity: without taste, every AI‑generated product looks alike, leading to a tsunami of indistinguishable apps.

5. Embedding Taste into the Product Process

Move from “human‑in‑the‑loop” to “human‑on‑the‑loop”: encode clear judgment rules into design systems, CI checks, and AI prompts. Example rule: “All user‑facing error messages must include a next‑step suggestion.”

Adopt a three‑layer translation:

Intuition → verbal judgment (e.g., “the copy is too long”).

Judgment → quantifiable rule (e.g., “titles ≤ 8 characters”).

Rule → automated constraint (e.g., lint rule in CI).

Only when taste reaches the third layer does it become scalable.

6. Deeper Layers: Generational Dialects and Meta‑Ability

“Taste” varies across generations and regions—what feels premium to Gen‑Z may seem gaudy to older users. The core is precise empathy for the target segment.

Beyond static taste lies a meta‑ability: sensitivity to change. Markets shift; yesterday’s “conversational UI” may become a nuisance today. Continuous dissection, user dialogue, and calibration keep taste dynamic.

7. Bottom Line

In the AI era, execution alone won’t win; the decisive factor is the ability to judge what to build, what to cut, and how to align with user value. Cultivate taste through massive analysis, disciplined deletion, structured libraries, and data‑driven calibration, and encode it into the product system for lasting competitive advantage.