The Twenty Minute VCMikey Shulman, CEO @Suno: The Future of Music, What is Gonna Happen? | E1244
CHAPTERS
- 0:00 – 0:43
Why AI products need better interfaces than an empty text box
Mikey opens with a critique of the ChatGPT-style blank prompt box becoming the default UI for every AI product. He argues most AI experiences should be product-led, with the model becoming an implementation detail as releases shift from “model versions” to “product versions.”
- •Empty text boxes work for ChatGPT but are the wrong interface for most AI products
- •The industry will stop shipping “model releases” and mostly ship product updates
- •Users will care about how music makes them feel, not what model generated it
- 0:43 – 2:12
What Suno is (and the origin: sense-making audio to generative music)
Mikey defines Suno as a tool that turns listeners into active music participants—creating, editing, and sharing. He explains the company initially expected audio to lag like early GPT-era text tools, but generative quality arrived sooner than expected, pulling them toward creation-first experiences.
- •Suno’s mission: “not making music, making musicians”
- •Early expectation: audio would be stuck in ‘understanding/sense-making’ mode for years
- •Generative audio improved faster than anticipated, reshaping the product direction
- 2:12 – 3:07
Scaling laws in music: why bigger isn’t always better
The conversation shifts to whether scaling laws will keep driving progress in music models. Mikey argues music is fundamentally subjective, so pure scale and benchmark-chasing are less decisive than techniques that produce “taste.”
- •Audio may be behind text, but the trajectory isn’t identical
- •Music optimization is subjective, unlike many objective text benchmarks
- •Models can remain relatively smaller; “taste” requires other methods
- 3:07 – 4:16
Music should feel like a video game: interactivity, engagement, and willingness to pay
Mikey explains his “music as a video game” analogy: interactive, social, and immersive experiences that people actively engage with. If music becomes more participatory and fun, he believes consumers will pay more—like they do for games.
- •Video games are interactive, rich, social experiences; music can be too
- •Today’s music is often passive ‘background’ consumption
- •More engagement could expand willingness to pay and grow the industry
- 4:16 – 5:27
From model versions to product versions: pricing, perception, and what powers Suno
Harry and Mikey agree that charging more for ‘better models’ is likely temporary; the model will fade from the user’s awareness. Mikey shares Suno uses a transformer and that differentiation comes from audio representation/tokenization rather than novel architecture.
- •Future: users won’t track model versions; they’ll experience a product
- •Suno uses transformer architecture
- •Key advantage: audio representation/tokenization choices
- •Borrowing scaling playbooks from open-source text while innovating on audio tokens
- 5:27 – 6:24
Economics of infinite supply: average value drops, total value rises
Harry challenges the idea that Suno increases music’s value despite expanding supply. Mikey argues the average value of a track may fall, but the overall societal and individual value of music rises via broader participation and richer experiences.
- •Infinite/near-infinite supply can reduce value per item
- •But it can greatly increase total engagement and industry vibrancy
- •Goal: expand the ‘participant’ base, not keep music precious and exclusive
- 6:24 – 8:27
Quantum computing: incredible promise, bad near-term career bet
Mikey explains why quantum computing is exciting but commercially premature: core problems remain physics and basic research. He suggests government or public-private partnerships are better suited than venture capital, and advises individuals to be cautious joining quantum startups now.
- •Quantum progress is still constrained by fundamental research
- •VC time horizons don’t match decades-long development cycles
- •Most quantum startups have struggled; public-private models may work better
- •Career advice: extremely high risk right now
- 8:27 – 9:41
Talent and experimentation: why economists and physicists do well in ML
Mikey describes recruiting as finding underappreciated talent and explains why economists and experimental physicists can excel. Economists bring first-principles thinking and natural experiments; physicists run fast, high-quality experiments—key in empirical ML.
- •AI company-building is largely a talent-finding game
- •Economists: strong at natural experiments and interpreting benchmarks
- •Experimental physicists: exceptional at fast, rigorous empirical iteration
- •Winning comes from running more high-quality experiments quickly
- 9:41 – 12:29
Competing with AI giants: location, mission, and aligning models to taste
Asked how Suno competes with OpenAI/Anthropic compensation, Mikey points to being Cambridge-based and tackling a unique problem: alignment to human taste rather than objective truth. He outlines how usage data and A/B testing help, while noting RLHF-style methods may not fully translate to music.
- •Suno competes via mission and problem uniqueness, not top-of-market pay
- •Cambridge HQ helps attract those who want that ecosystem
- •Core challenge: aligning to subjective taste, not objective correctness
- •Heavy use of A/B testing and preference-based techniques; future may be more personalized
- 12:29 – 17:32
Product truths: time-to-wow, paywall metrics, and why Suno charged from day one
They discuss Suno’s fast ‘time to wow’ and how even small added latency reduces user delight. Mikey defines a successful user as someone who hits the paywall on day one (even if they don’t pay), and argues charging early avoided novelty status while generating crucial segmentation data for product learning.
- •Latency matters: users like 8 seconds more than 10 seconds
- •Success metric: hitting the paywall day one as proof of sustained delight
- •Charging early signaled seriousness and avoided being a gimmick
- •Paywall data identifies who to interview and what moments drive subscription
- •GPU cost is the largest expense; revenues help offset compute burn
- 17:32 – 22:37
RIAA lawsuit and the path to AI–incumbent coexistence
Mikey cautiously acknowledges copyrighted works exist in training data, framing it as standard across AI and not inherently illegal. He laments a ‘fixed pie’ mindset in music and argues lawsuits divert resources from building a larger future; he advocates collaboration over “disrupt vs sue” extremes.
- •Copyrighted works appear in training data; common across AI companies
- •Music is highly litigious; lawsuits aren’t surprising but are costly
- •Critique of ‘fixed pie’ industry thinking; focus should be growing the pie
- •Preferred approach: talk and partner first, litigate last
- •If AI tools are forced away, the industry may miss a gaming-sized upside
- 22:37 – 29:02
One unified music world: creators, platforms, and artist participation
Mikey argues consumers won’t tolerate separate ecosystems for AI music vs traditional music; it should feel unified across creation and listening. He shares that many artists privately use Suno, and explores opt-in personalized artist models—where artists control and monetize fan-driven creation, akin to fan fiction.
- •Two separate ‘worlds’ (AI vs normal music) create friction for consumers
- •Future likely blends creation + consumption experiences
- •Many artists privately admit they use and like Suno
- •Opt-in personalized artist models could enable new revenue and engagement
- •Suno restricts impersonation today; authorized use changes the ethics and economics
- 29:02 – 33:17
New business models and industry partnerships: Timbaland, remix culture, and boredom in pop
Mikey describes how paying to create (not just listen) opens new monetization paths beyond stream-share economics. He highlights a Timbaland remix contest as a model for deeper fan engagement, explains the partnership (advisor/equity, product feedback loop), and critiques pop’s growing homogeneity driven by production tech and platform incentives like TikTok.
- •Creation-based monetization: users pay to make music, not only to stream it
- •Opt-in ‘forks’/Patreon-like direct support and remix rights as new models
- •Timbaland partnership: product collaboration + legitimacy + creator ‘cover’
- •Remixing idols can be a deeper engagement than backstage access
- •Pop music trends: shorter, more homogeneous structures shaped by streaming/TikTok incentives
- 33:17 – 36:26
Discovery in an era of infinite music: Spotify, TikTok, YouTube, and engagement wars
They explore how algorithmic discovery already is, and how infinite supply changes the stakes. Mikey discusses platform roles—TikTok breaking songs, Spotify recommending, YouTube’s engagement advantage—and why Suno refuses to be merely background audio for AI video tools, aiming instead to increase music’s intrinsic value.
- •Discovery is already heavily algorithmic; people underestimate its influence
- •TikTok often drives what becomes popular elsewhere
- •YouTube is more engaging; Spotify is pushing video to compete
- •Suno rejects API requests from AI video companies to avoid ‘background music’ commoditization
- •Goal: make music more engaging than passive background listening
- 36:26 – 40:22
Founder regrets and product philosophy: leaving Discord, great UI, and moving beyond ‘prompts’
Mikey’s biggest regret is staying on Discord too long; a thin web app rapidly took most traffic, proving UI’s importance. He defines great UI as enabling workflows that DAWs can’t—like genre-transforming a whole song—and hopes the industry stops talking about ‘prompting’ in favor of more natural interaction paradigms.
- •Regret: assumed Discord could be permanent; web UI changed everything
- •Discord remains valuable as community, but not as primary product surface
- •Great UI enables new workflows, not prettier versions of old ones
- •Example workflow: transform a full song into a different genre as a first-class feature
- •Prompt guides are a symptom; the future should feel less like ‘prompting’ and more like being understood
- 40:22 – 45:40
Scaling the company: remote-work exceptions, capital as a weapon, and balancing growth with revenue
Mikey debates remote work: exceptions can be worth it but become hard to manage at scale, requiring judgment. He discusses raising capital quickly to create step-changes in capability (especially compute), notes subtle VC dynamics around pro-rata, and explains Suno’s revenue success as simply prioritizing monetization alongside growth without treating them as mutually exclusive.
- •Remote-work exceptions scale poorly; judgment becomes the bottleneck
- •GPU spend dominates costs; capital enables step-function improvements
- •VC ‘reserves for later’ often means firm pro-rata expectations
- •Revenue scaled because Suno cared about charging and tracked it from day one
- •Growth vs revenue isn’t yet a forced trade-off given how early the market is
- 45:40 – 1:00:26
Fear vs optimism about rapid change: dystopias, personalization tension, and building the ‘should’ future
Harry shares unease about rapid societal and technological change; Mikey stays optimistic but rejects ‘AI is inevitable’ complacency. He names two dystopias—unpermitted artist impersonation and hyper-personalized antisocial music-as-drug—and argues the future must be actively built toward social, participatory creation that expands who can make a living in music, similar to Instagram’s impact on photography.
- •‘AI is inevitable’ is a dangerous excuse for inaction
- •Dystopia 1: mass artist impersonation without permission or payment
- •Dystopia 2: hyper-personalized, antisocial ‘music drip’ optimized to your brain
- •Good personalization is social (e.g., making a song for someone else)
- •Desired future: more people creating for more hours; economics follow increased engagement
- •Analogy: Instagram expanded who can earn from photography; AI can do similar for music
