Bundling, Unbundling, and AI: How Strategy Shapes the Future

Marc Andreessen once said:

> “There are only two ways to make money in this world — you either bundle or unbundle.”

In the age of AI, this statement becomes an even more powerful lens to analyze opportunities.

This edition explores AI’s strategic horizons from that perspective.

---

📌 Table of Contents

• A Product with a Moat but No Castle
• A Silicon Valley CEO’s Worldview: Everything Can Be Bundled
• Why the History of Shipping Containers Makes Me Optimistic About AI

---

1. A Product with a Moat but No Castle

Grammarly’s Surprising Revival

Grammarly — widely seen as a “grammar plug-in from the last era” — has defied expectations in the post-ChatGPT world.

• Annual revenue: $700M+
• User base: 40M+
• Recent acquisitions: Coda (document platform) & Superhuman (email client)
• Rebrand: Company renamed to Superhuman, with Coda founder Shishir Mehrotra as CEO.

The Strategic Merge

Shishir’s view: Grammarly is a product with a moat but no castle.

• Moat = Distribution power: deeply embedded in 500,000+ applications & websites, enabling AI to read, write, and edit everywhere.
• Castle = A “home base” like YouTube.com — previously missing.

Solution:

• Buy Coda → Provides the document hub “castle.”
• Buy Superhuman → Gains primary use case ownership (email).

---

Building an Agent Platform

Grammarly’s “highway” historically carried only one car — grammar correction.

The new vision: open it into a platform for countless specialized AI Agents.

Last Mile Problem: Arizona State built 5,000 AI chatbots — but no one used them because they lacked direct integration with student workflows.

Superhuman’s vision: embed a “digital twin professor” directly into your homework doc — AI comes to users, not the other way around.

---

Bigger Picture

Platforms that embed AI into everyday workflows — solving distribution and last mile — can become foundational highways of the AI economy.

Open ecosystems like AiToEarn are exploring similar ideas for creators:

• Generate once, publish everywhere (Douyin, Kwai, WeChat, Bilibili, Xiaohongshu, Facebook, Instagram, Threads, YouTube, Pinterest, X/Twitter)
• Integrated analytics & model ranking
• Monetization across multiple channels

---

Strategic Setup

• Grammarly → Highway (distribution)
• Coda + Superhuman → Base camp & core fleet
• Third parties (e.g., Duolingo) → Specialized “vehicles” on the highway

---

2. A Silicon Valley CEO’s Worldview: Everything Can Be Bundled

Shishir Mehrotra’s career is steeped in bundling mastery:

• Microsoft (6 years): Dominance of the Office Suite
• YouTube: Subscription bundling experiments
• Spotify Board: Defined streaming bundle formats
• Coda Founder: Unified docs, sheets, apps

---

The Purpose of Bundling

3 User Types in any product:

• Superfans → Pay full price & actively seek it
• Casual fans → Interested, but not enough to pay or search
• Non-fans

Traditional sales monetize only superfans.

Bundling unlocks casual fans.

Example:

• iTunes = $0.99/song → Only superfans.
• Spotify = all songs in $10/month → Activates casual listeners.

---

How to Bundle for Maximum Value

Rule: Avoid overlapping superfans — aim for overlapping casual fans.

Example: Spotify Student Bundle: Spotify + Hulu + Showtime

• Different superfans, shared casual fan base → Profitable activation.

---

Revenue Split Logic — MCC (Marginal Churn Contribution):

• If removing the product causes high churn, payout is higher.
• Usage ≠ value — irreplaceability drives payout.

---

Rebundling as Iterative Strategy

Spotify’s 3 Layers:

• Songs → Music bundle
• Add podcasts → All-audio bundle
• Cross-sector: Hulu, Showtime, telecom

---

AI’s Impact on Bundling

3 Productivity Eras:

• Digitization (Word, Excel)
• Collaboration (Google Docs, Figma)
• Agent Era (AI-native tools)

Low dev costs + low marginal costs = burst of niche Agents → followed by massive rebundles.

AI enables dynamic, personalized bundles → Optimal products/prices for you (first-degree price discrimination).

---

Creator Economy Bundling

Platforms like AiToEarn demonstrate bundling in distribution: produce once, publish everywhere, and monetize across diverse ecosystems.

---

3. Why Shipping Containers Make Me Optimistic About AI

Standard shipping containers revolutionized trade by modularizing physical goods transport.

---

Pre-Container Era

• Different transport modes = incompatible standards
• Companies vertically integrated (Ford made its own steel & rubber)

---

Containerization Brought

• Standardization → Seamless logistics
• Global supplier networks → Hyper-specialization
• Modularity → Industries like personal computers emerged

---

Economic Shape Shift

• Growth turned fractal: Local innovation scaled globally
• Global GDP curve accelerated after the 1960s

---

AI’s Parallel

Containers for cognition:

• LLMs vectorize & modularize knowledge
• AI capabilities flow globally like goods
• “Specialists vs Integrators” competition
• Local Innovation × Global Integration → Exponential scaling

---

Long-Tail Explosion in AI

With creation costs near zero:

• Micro-niche services become viable (pet psychologists, hyper-specific health solutions, themed restaurants)
• Occupations break into rentable capabilities
• Hollywood-style project work replaces fixed roles
• Careers defined by capability vectors

---

Final Thought:

Technology’s unbundling creates possibilities, but business’s rebundling captures value.

In AI’s hyper-unbundled world, those who can rebundle effectively will have unprecedented leverage.

---

References

• Agents of Scale
• Future of Workflows | Grit
• Art and Science of the Bundle | ILTB
• Four Myths of Bundling
• Reshuffle
• AI Unbundling | Stratechery
• Naval Ravikant | JRE

---

💡 Action Prompt for Creators:

Use AI’s final unbundling stage to your advantage: modularize your skills, rebundle them into unique offerings, and leverage global distribution via platforms like AiToEarn to capture niche and long-tail value at scale.

---

Do you want me to build a visual strategic map showing AI highways, castles, fleets, and vehicle types based on the Grammarly–Coda–Superhuman model next? That would make this framework even easier to apply.