Overview of 20VC: Why Now Is the Time for the Application Layer
In this episode of 20VC, Harry Stebbings sits down with Mike Mignano, newly appointed GP at Union Square Ventures (USV), former founder of Anchor, and early investor in companies like Suno and Granola. The conversation centers on the AI stack’s next phase: after the massive infrastructure build-out, the opportunity has shifted toward applications, specialized workflows, and products that are deeply aligned with users’ goals. Mignano argues that model providers will remain powerful, but they will not automatically own the app layer—especially where startups can move faster, build more context, and solve hard, vertical problems. The episode also covers venture strategy in the AI era, token spend, routing layers, open vs. closed models, energy as the hidden constraint on AI, and why media is increasingly moving toward self-publishing and independent platforms.
The Core AI Thesis: Infrastructure Is Built, Now Applications Win
Mignano’s central view is that AI is moving from infrastructure creation to application creation.
- The model and infrastructure layer has seen enormous capital investment and technological progress.
- Now that the underlying “new toys” exist, there’s room for a broad application layer to emerge.
- He compares this to the early internet, when fiber and broadband came first, followed by the app layer that actually transformed usage.
Why the App Layer Matters Now
- Applications will capture value by being:
- Fast to market
- Highly contextual
- Deeply integrated into user workflows
- Clear about the job-to-be-done
- He believes the best companies will be built by founders who know exactly what they’re looking for and can move quickly.
Open vs. Closed Models, Harnesses, and Routing Layers
A major topic is whether model companies like OpenAI and Anthropic will dominate the app layer. Mignano’s answer: not entirely.
Open vs. Closed
- In a more mature AI market, enterprises will increasingly optimize for:
- Cost
- Model capability
- Task-specific efficiency
- That means more use of:
- Open-weight and open-source models
- Routing layers that send tasks to the best model for the job
- Human-aligned “harnesses” built around models
What Is a “Harness”?
Mignano describes a harness as the application layer tightly coupled to the model—for example:
- Claude Desktop / Claude Code
- Products like Hermes
- Products like Pi
These are products that shape how the model is used, and increasingly determine whether the user experience feels aligned and valuable.
Why Routing Matters
- Routing layers can optimize spend by selecting the cheapest or best model for each task.
- He sees room for real businesses here, though it’s hard to build a giant standalone routing company unless it becomes deeply embedded in developer workflows.
- One possible model: a bounty system where routing earns fees for choosing the best or most efficient model.
TokenMaxxing: Why Startups Should Still Spend Aggressively on Tokens
One of the clearest takeaways is Mignano’s view on token spend.
For Startups
- Startups should maximize token spend on the right tasks, especially coding.
- The logic: startups need every advantage possible.
- In coding and high-leverage work, frontier models can provide a meaningful edge over incumbents.
For Big Companies
- Large incumbents may need to constrain usage because token spend at scale can damage economics.
- He points to the tension between employee productivity and cost control at companies like Salesforce or Microsoft.
His Bottom Line
- Startups should use the best available tools to win.
- Smaller, mission-driven teams can more tightly manage spend while still moving faster than large enterprises.
Why USV Is Betting on “Obliteration,” Not Automation
Mignano explains that USV’s philosophy is to back companies that obliterate markets rather than merely automate existing workflows.
Automate vs. Obliterate
- Automate = improve current processes.
- Obliterate = reinvent the category or remove the old business model entirely.
Examples
- Doctronic: not just making medical practices more efficient, but putting a doctor in everyone’s pocket.
- Suno: not just improving music tools, but creating a new behavior around “creative entertainment.”
This is a key distinction in how USV thinks about venture returns and category creation.
Where Model Providers Can Still Win—and Where They Probably Won’t
Mignano is not dismissive of model providers. In fact, he believes they remain formidable. But he thinks they won’t swallow everything.
Why They Might Win in Some Cases
- They have the infrastructure
- They have compute advantages
- They are often at the frontier
- They have massive data and distribution leverage
Why They Won’t Win Everything
- Enterprises still care about trust, context, and specialization
- Regulated markets are hard to enter quickly
- Strong startups can build real moats through:
- Distribution
- Integration
- Compliance
- Workflow ownership
- Context accumulation
The Granola Example
Mignano notes that products like Granola benefit from:
- Focus on one core use case
- Deep context inside organizations
- A workflow that becomes hard to rip out once adopted
He also acknowledges the threat from bundling, especially from Microsoft and other incumbents.
The Future of AI: Recursive Self-Improvement vs. Commodity Plateau
Mignano lays out two possible long-term futures for AI:
1) Exponential Takeoff
- A lab reaches recursive self-improvement
- AI can improve its own research and capabilities
- The first mover could run away with the market
2) A Plateau
- AI hits a technical or practical limit
- The market becomes more commoditized
- Competition shifts to:
- Price
- UX
- Routing
- Workflow integration
- Trust and alignment
His view: either path is plausible, but he leans toward the idea that the ecosystem will remain competitive rather than being completely captured by one or two players.
Energy Is the Hidden Layer Beneath AI
A big part of the conversation turns to energy, which USV has been betting on for years.
Why Energy Matters
- AI requires enormous compute
- Compute requires power
- Power requires infrastructure and portability
Investment Themes Mentioned
- Fuse Energy
- Radiant: small nuclear reactors
- Rune: micro data centers near generators and wind farms
- Panthalassa: data centers at sea
His point: if AI is going to scale, the energy stack becomes a crucial source of innovation and venture opportunity.
Media Is Being Unbundled
Mignano is bullish on independent media and skeptical of traditional media.
His View
- Traditional media is in decline
- Independent creators and self-publishing platforms are winning
- The real value lies in editorial freedom and direct audience relationships
Platforms He Likes
- X
- YouTube
- Spotify
- Substack
He views Substack as one of the best examples of enabling creator independence and monetization.
Venture Investing in the AI Era: What Changes?
Mignano shares a strong view that venture itself is changing.
Small Funds vs. Large Funds
- He believes very large platform funds can produce venture-like returns in AI because they can participate in capital-intensive frontier companies.
- But small funds still have an edge in:
- Opinionated investing
- Early ownership
- High-conviction bets
- Specialized theses
Ownership and Price
- At seed and Series A, ownership still matters a lot.
- At later stages, if the upside is large enough, ownership matters less than getting in at all.
- He says price should be used as a litmus test for conviction.
His Advice
- Don’t pass on price too easily, but don’t ignore fund math.
- Be willing to pay more when conviction is high.
- In the early days, focus on exceptional founders and market size.
Biggest Lessons From Venture and Founding
Mignano reflects on what he has learned as both founder and investor.
1) Don’t Project Your Ideas Onto the Founder
- Even if you’re right, it may not be the founder’s plan.
- Great investors back founder judgment, not just their own opinions.
2) Founder Matters Most
He says his ordering has changed over time:
- Earlier: product > market > founder
- Now: founder > market > product
Why?
- Founders pivot
- Execution matters
- Resilience matters
- Communication matters
3) Communication Is a Huge Filter
One of the biggest mistakes he says he’s made in evaluating founders is underestimating communication ability.
Communication affects:
- Recruiting
- Fundraising
- Team alignment
- Product vision
- Market storytelling
His Biggest Misses and Hits
Misses
- Suno: passed initially due to ownership concerns
- Granola: underestimated the founder and the idea’s potential
- Substack: regrets not investing and now sees it as a defining platform for independent publishing
Hits / Strong Convictions
- Suno: later recognized as a thesis + founder fit
- Granola: a strong founder bet on Chris
- Bored: his best first founder meeting, in his view
What Excites Him Most About the Next 5–10 Years
Mignano ends on a highly optimistic note.
He wants:
- To work with people he genuinely enjoys
- To partner with exceptional founders
- To generate generational fund returns
For him, fun and fulfillment are not separate from performance—they’re part of what makes strong venture outcomes possible.
Key Takeaways
- The AI infrastructure build-out is largely done; the next wave is application-layer value.
- Model providers are powerful, but they won’t automatically own every product category.
- Startups should “tokenmaxx” on the right tasks, especially coding.
- The best AI businesses will likely be focused, highly contextual, and hard to rip out.
- Routing layers and harnesses may become important infrastructure for the app ecosystem.
- Energy is a major bottleneck and a major investment opportunity.
- Traditional media is weakening while independent media and self-publishing continue to grow.
- In venture, founder quality, speed, and conviction matter more than ever.
Notable Quote
“Don’t automate, obliterate.”
That line captures the USV worldview in this episode: build something that doesn’t just make the old world faster, but changes the structure of the market itself.