ExpertsFMSuno's Mikey Shulman: Everyone Can Make Music Now
Training Data · May 13, 2026
5 Selected Lessons:
heuristic · 2 framework · 1 priority · 1 reasoning_pattern · 1
Collaboration over adversarial disruption with incumbents
Source
Training Data — Suno's Mikey Shulman: Everyone Can Make Music Now (May 13, 2026) · at 24:12 ▸
When this applies
when your technology disrupts an existing industry with powerful incumbents, prioritize building positive-sum products together over trying to replace or antagonize them
The takeaway
When disrupting an industry with entrenched incumbents who control critical complementary assets (culture, rights, relationships), prioritize partnership structures that create new joint value over pure substitution strategies.
Why it works
Incumbents hold cultural capital, relationships, and rights that a disruptor can't replicate; co-building new products that use both sides' strengths creates value unavailable to either alone, converting adversaries into partners.
One level up
Positive-sum integration with incumbents is often more durable and valuable than zero-sum disruption, especially when the incumbent controls assets the disruptor cannot replicate.
Where it stops holding
This stops working when the incumbent's interests are so fundamentally opposed (e.g., existential threat with no possible shared upside) that no positive-sum structure is achievable.
Detail
hierarchy:
- Avoid creating two separate worlds (AI music vs non-AI music) since most music will have some AI in it
- Build new joint products with labels that couldn't have existed before
- Deepen artist-fan connection as the shared value creation target
- Monetize in a way that benefits artists, fans, and rights holders simultaneously
rationale: Record labels are culturally important institutions that cultivate stars; working with them to build new interactive products benefits all parties, versus framing the relationship as adversarial.
In their words
What I'm most excited about doing with Warner is actually building things together that could never have existed before and building products that let fans interact with their favorite artist and really deepen the artist-fan connection in ways that are just positive sum for everyone.
Arbitrary release cadence to avoid research paralysis
Source
Training Data — Suno's Mikey Shulman: Everyone Can Make Music Now (May 13, 2026) · at 10:11 ▸
When this applies
when running continuous R&D with no natural stopping point, impose an artificial regular release cadence rather than waiting for a 'perfect' version
The takeaway
Impose artificial, regular shipping checkpoints on an open-ended improvement process to guarantee continuous delivery of value instead of waiting for completeness.
Why it works
Continuous research has no natural endpoint; without a forced cutoff, teams risk indefinite delay chasing an ever-receding ideal.
One level up
Discipline around cadence protects against the failure mode of infinite perfectionism in any ongoing improvement process.
Where it stops holding
This stops working when shipping too frequently causes instability or when a genuinely discontinuous breakthrough requires withholding release until mature.
Detail
procedure:
- Maintain an aggressive, continuous research roadmap
- Pick an arbitrary cutoff point to bundle finished improvements into a release
- Push everything else to the next version
- Keep a steady release cadence
benchmarks: {}
rationale: Avoids the failure mode of not releasing for years while chasing a perfect model.
In their words
Because what you would hate to have happen is we don't release stuff for like two years and we try to make, you know, the music model to save humanity.
Mikey's stated preference for 'never possible before' bets over automating the familiar
Source
Training Data — Suno's Mikey Shulman: Everyone Can Make Music Now (May 13, 2026) · at 32:46 ▸
When this applies
when choosing between building something that automates an existing well-understood process versus something that creates an entirely new category of behavior, weigh the motivational and differentiation value of doing something previously impossible
The takeaway
Mikey says he and his team favor pursuing capabilities that were previously impossible over optimizing known business processes, believing novelty creates differentiation and intrinsic motivation, while still acknowledging that automation has real business value.
Why it works
In Mikey's view, automating known processes has clear but bounded value and invites direct competition from others optimizing toward the same known target, whereas building something previously impossible creates both differentiation and stronger motivation for him and his team.
Where it stops holding
This stops working when the automation opportunity is large, clearly monetizable, and low-risk, while the 'newly possible' bet has unclear demand — pure novelty-seeking isn't inherently superior economically.
Detail
approach: Mikey asks whether a proposed AI application is automating something that already exists or creating something that couldn't have existed before, and says he weights the latter more heavily for personal/team motivation.
signals:
- he believes it takes real creativity to see how AI can seep into play and creation, not just automate business processes
- he observes most people default to imagining 'AI Spotify' rather than new participatory formats
rationale: He explicitly says he and his team are 'much more inspired and motivated by doing something that wasn't possible until today instead of automating or speeding up something that already exists.'
In their words
I think we are much more inspired and motivated by doing something that wasn't possible until today instead of automating or speeding up something that already exists.
Model the raw substrate, not human abstractions
Source
Training Data — Suno's Mikey Shulman: Everyone Can Make Music Now (May 13, 2026) · at 0:00 ▸ · at 7:17 ▸
When this applies
when building any AI system, avoid baking in the domain's existing categorical abstractions (e.g. fixed taxonomies) because they cap what the model can ever produce
The takeaway
When modeling a domain, prefer representing the underlying continuous reality over the human-invented discrete categories layered on top of it, because the categories are a ceiling, not a feature.
Why it works
Encoding a human-made abstraction (like 12 discrete tones) as a hard constraint in the model's representation space permanently limits the output distribution to that abstraction's boundaries.
One level up
Removing human-imposed constraints from a system's internal representation expands its creative/combinatorial ceiling at the cost of more modeling difficulty.
Where it stops holding
This stops working when the discrete abstraction actually reflects a hard physical or perceptual limit (not an arbitrary human convention), in which case modeling the raw substrate adds complexity without expanding real capability.
Detail
rule: Model the continuous raw signal, not the discrete human taxonomy on top of it.
threshold: If you give the model 12 tones or 200 instruments, those become hard caps on output.
rationale: The model only ever outputs what its representation can express; discrete constraints become permanent ceilings.
In their words
In Western music, there are 12 tones. If you tell the model there are 12 tones, it will only ever produce those 12 tones. You will be forever limited.
Model quality alone is not durable moat; invest in product/UX
Source
Training Data — Suno's Mikey Shulman: Everyone Can Make Music Now (May 13, 2026) · at 27:22 ▸
When this applies
when a well-resourced competitor can eventually match your underlying model/technology, shift competitive investment toward product experience rather than relying on model superiority alone
The takeaway
In markets where a deep-pocketed competitor can match your core technology given enough spend, build durable advantage through product experience and delight rather than model capability alone.
Why it works
Well-capitalized competitors can out-invest in raw model capability over time, eroding a pure-model advantage, whereas product/UX quality compounds via culture, taste, and iteration that's harder to directly copy.
One level up
Where technical capability is catchable by capital, shift the locus of competitive advantage to execution and experience, which is harder to buy.
Where it stops holding
This stops working when the underlying model capability gap is so large or defensible (e.g., proprietary data/algorithms) that competitors genuinely cannot catch up regardless of spend.
Detail
rule: Don't over-rely on model quality as moat; invest heavily in UX/product delight.
threshold: A deep-pocketed competitor (e.g., Google) 'will outspend us seven days a week and can probably catch up on the model side.'
rationale: The average consumer won't tolerate rough edges since they're paying for fun, not work, so product polish matters more in consumer AI than in enterprise.
In their words
It's unclear how much moat exists in only a model. And it's like, I'll just say it, like Google has started to build music models. And while ours are way better today, they're Google and they'll outspend us seven days a week and they can probably catch up on the model side.