Skip to content

Debates

Is taste a decisive, cultivable capability in building products?

12 recorded positions from 8 people, first said Nov 15, 2024. They do not agree — the readings below are what each one actually argued.

Ai handles drafts but humans supply taste and point of view

Victor Riparbelli · Jan 15, 2025

Removing technical barriers does not remove the need for skill — storytelling, taste and knowing what's important will matter more than ever, while distribution connections and deep visual-effects craft will matter less

AI can automate the technical layers but not the judgment about what to say and how to combine familiar and novel elements so content stands out

Scope: concedes Harry's point about skill requirements

40:38 20VC: Why Scaling Laws Will Not Continue | OpenAI vs Anthropic vs X.ai: Who Wins and Why | How Far Will Model Providers Go Into the Application Layer | The End State for Models: Many Specialised or Few Generalised with Victor Riparbelli @ Synthesia

Kieran Flanagan · Jul 11, 2025

AI is a great assistant for media production — first drafts of video and scripts — but usually cannot produce genuinely tasteful content; a human is still needed to inject taste, personality and point of view

AI handles the draft layer well but taste and point of view still require people

Scope: 'maybe in some respects' AI can be tasteful; most of the time not

49:42 20Growth: The Death of Growth Teams? | How Hubspot Use AI to Triple Email Conversion | The Future of AI SEO | Why Prompt Engineering is the New Coding | What Every CMO Needs to Know About AI in 2025

Anton Osika · Aug 18, 2025

The current slow, detailed, one-person design process will be replaced by humans specifying design philosophy at a high level while AI does the implementation and humans feed back external context

Humans are too obsessed with small details being perfect, which makes them move much slower; an opinionated AI pipeline can carry a product all the way through marketing, growth, testing and QA

27:40 20VC: Lovable CEO Anton Osika on $120M in ARR in 7 Months | The Honest Truth About Defensibility and Unit Economics for AI Startups | The State of Foundation Models: Long Grok, Short OpenAI, Why | Replit vs Lovable vs Bolt: What Happens

Execution speed and taste trade off and must match the opportunity

Nabeel Hyatt · Apr 4, 2025

Execution speed and taste are directly in conflict in founders, and the right point on that spectrum must be cast to match the specific startup opportunity

Pure shoot-first founders aren't introspective and chase shiny pennies, while founders who are all taste never ship; highly competitive markets with a clear roadmap demand top-0.0005% execution, while creating a new market demands incredible taste

26:44 20VC: Why To Win in AI, Investors Need to Change Their Approach | Why VC is Run by Principals and Associates and is a Broken System | The Bull Case for Anthropic & Whether Deepseek Changes Their Strategy with Nabeel Hyatt @ Spark Capital

Nabeel Hyatt · Apr 4, 2025

The best founders can tell at any given moment which muscle — execution or taste — they should be using and how to use it

30:14 20VC: Why To Win in AI, Investors Need to Change Their Approach | Why VC is Run by Principals and Associates and is a Broken System | The Bull Case for Anthropic & Whether Deepseek Changes Their Strategy with Nabeel Hyatt @ Spark Capital

Also on the record

Raman Malik · Nov 15, 2024

Taste is the ability to recognise what good looks like and reverse-engineer it, and that skill matters more than being excellent at any individual craft

He believes he isn't actually that good at anything himself; his edge is observing what's working in the world and reverse-engineering it

45:58 Taste is recognizing and reverse engineering what good looks like not craft excellence

Raman Malik · Nov 15, 2024

You detect taste in an interview by the depth of the questions a candidate asks about your actual problems, not by testing them on artifacts

Being an open book about company challenges and jamming on them reveals very quickly how deeply someone is thinking and whether they can reverse-engineer good efficiently

46:21 Depth of questions about real problems reveals taste not artifact testing

Shiv Rao · May 16, 2026

Taste is not just judgment — it is seeing patterns and combining things in ways that feel different and authentic, and it is what distinguishes the truly magical companies today.

In the best products you can feel the human behind it, the decisions they made and the things they said no to.

9:35 Taste is pattern combination and the key differentiator

Shiv Rao · May 16, 2026

Cultivating taste requires deliberate exposure — reading the latest ML papers, tracking new UI/UX patterns, living at the edge of culture — while still being discerning rather than tasting everything.

If you don't expose yourself to certain flavors you won't even be aware they exist.

10:26 Taste is cultivated by deliberate discerning exposure

Max Junestrand · Aug 15, 2025

Easy LLM prototyping will not let companies skip the design stage; end-to-end experience design remains necessary

Without a design eye owning the whole experience you get a Frankenstein product where individual pages look fine but the overall experience is incoherent, and design has to be integrated with brand

73:12 Easy llm prototyping does not eliminate need for end to end design

Paul Erlanger · Jun 27, 2026 · hedged

Design becomes increasingly important relative to engineering as AI absorbs mid-level engineering tasks

Larger companies have many mid-level engineers doing work AI can take over fairly easily

29:01 Design outweighs engineering as ai absorbs mid level work

Nabeel Hyatt · Apr 4, 2025

Granola's Chris demonstrated real taste by completely resetting the product between the seed and Series A despite okay internal metrics, where competitors would have built a me-too product and tried to out-execute

He could self-inspect and recognize it wasn't working even though the metrics didn't force it

29:30 Resetting product despite ok metrics demonstrates real taste

Your assistant can query this graph directly — 12 positions here, 19,646 across the corpus. Add 996.fm over MCP.