Platform Divergence, Scale Gaps, And Ecosystem Differentiation
Sources: 1 • Confidence: High • Updated: 2026-03-11 09:12
Key takeaways
- Claude and ChatGPT app directories each have more than 200 apps, with only about 11% overlap between them.
- As people use AI for both personal and professional topics, misapplied memory can feel jarring, creating a need to model user identity and context boundaries correctly.
- A Pew Research study is reported to find that over half of teenagers use AI for homework, with 38% using it as a creative tool, 16% for casual conversation, and 12% for emotional support and advice.
- Consumer AI usage is expanding beyond chat prompt boxes into AI browsers and desktop applications such as Cursor and Granola.
- OpenClaw is reported to have surged in technical adoption (including becoming the most-starred GitHub project and being estimated to debut around #30 on a web list based on February data) while sign-up interest plateaued and remained concentrated among technical users.
Sections
Platform Divergence, Scale Gaps, And Ecosystem Differentiation
- Claude and ChatGPT app directories each have more than 200 apps, with only about 11% overlap between them.
- ChatGPT usage is reported to substantially exceed Gemini and Claude, at about 2.7x Gemini on web and 2.5x on mobile, and roughly 30x Claude on web and 80x on mobile.
- Major consumer AI platforms are diverging in strategy: Claude is positioned toward prosumer tools, Gemini usage tracking is tied to creative model releases, and ChatGPT is evolving toward an 'everything app' with ads, transactions, and subscriptions.
- Claude’s app ecosystem is described as skewed toward premium data and research tools, while ChatGPT’s is described as skewed toward consumer marketplaces such as travel, nutrition, and consumer finance.
- Lock-in for horizontal AI assistants is expected to increase via network effects, developer focus on the largest user base, and a potential 'log in with ChatGPT' layer that carries memory and tokens across apps.
- Personal/work identity mixing and memory sharing may limit lock-in unless AI products can segment memory across different user personas.
Memory/Personalization As A Retention Driver And A Trust Boundary Problem
- As people use AI for both personal and professional topics, misapplied memory can feel jarring, creating a need to model user identity and context boundaries correctly.
- Lock-in for horizontal AI assistants is expected to increase via network effects, developer focus on the largest user base, and a potential 'log in with ChatGPT' layer that carries memory and tokens across apps.
- Personal/work identity mixing and memory sharing may limit lock-in unless AI products can segment memory across different user personas.
- Claude and ChatGPT are described as particularly good at memory, and Google is reported to have launched 'Personal Intelligence' that uses data from docs and email to personalize AI across apps.
- Within a couple of years, AI products that do not immediately feel like they know the user are expected to be perceived as broken and traditional onboarding is expected to largely disappear.
- AI systems are reported to provide materially higher value after two to three months of interaction than at initial use, consistent with benefits from accumulated memory and personalization.
Adoption Lag And Current Consumer Use Patterns
- A Pew Research study is reported to find that over half of teenagers use AI for homework, with 38% using it as a creative tool, 16% for casual conversation, and 12% for emotional support and advice.
- Cultural adoption of generative AI lags technological capability, with early-adopter behaviors reaching the mainstream after a lag of roughly six months.
- ChatGPT is described as the largest AI product globally but is used weekly by roughly 10% of the global population.
- At-scale consumer use of ChatGPT is reported to remain concentrated in homework-style and Google-like queries, with some companionship usage.
- Homework, creative tooling, casual conversation, and emotional-support uses of AI are expected to converge toward near-universal adoption over time.
Surface-Area Shift: Desktop Apps, Browsers, And Switching Costs
- Consumer AI usage is expanding beyond chat prompt boxes into AI browsers and desktop applications such as Cursor and Granola.
- An AI-native browser is expected to be a powerful consumer surface because it makes AI always-on and ambient where users already spend time online.
- Mainstream adoption of a new browser is constrained by switching costs from existing workflows, so AI browsers need clear killer features beyond feature parity.
- OpenAI released Atlas and Anthropic released Cowork.
Agents: Capability Shift And Defensibility Concerns
- OpenClaw is reported to have surged in technical adoption (including becoming the most-starred GitHub project and being estimated to debut around #30 on a web list based on February data) while sign-up interest plateaued and remained concentrated among technical users.
- Horizontal consumer agents are expected to be hard to defend as agentic capability commoditizes, making distribution from large platforms a key advantage versus standalone startups.
- Agents are expected to evolve from a distinct product category into a default capability of most AI and tech companies because they deliver outcomes rather than just inputs.
- Agents are expected to unlock consumer use cases such as finance, healthcare, travel planning, and complex shopping by reliably gathering and acting on data across systems.
Watchlist
- India is described as a high-potential AI market where many languages are poorly supported by current LLM and voice products, creating room for localized entrants.
Unknowns
- What are the underlying data sources, definitions, and time windows for platform usage comparisons and penetration estimates (weekly global usage, web vs mobile ratios, and country shares)?
- How stable is the reported ~11% app-directory overlap over time, and what categories drive changes in overlap?
- To what extent are ads and transactions actually being implemented in ChatGPT’s consumer surfaces, and what are the effective take rates or revenue mix versus subscriptions?
- Do memory systems measurably improve retention and user value across broad cohorts, and what is the magnitude of the lift (beyond anecdotal reports)?
- What product mechanisms effectively prevent misapplied memory or context leakage between personal and professional use (e.g., scoped profiles), and how do users respond?