The AI Daily Brief: Artificial Intelligence News and Analysis

Episode: What People Are Actually Using AI For Right Now
Host: Nathaniel Whittemore (NLW)
Date: December 8, 2025

Episode Overview

In this episode, Nathaniel Whittemore explores a data-driven snapshot of real-world AI usage, grounded in a recent study by OpenRouter and a16z analyzing over 100 trillion tokens of LLM interaction. Whittemore breaks down how developers, startups, and power users are actually deploying and engaging with AI, highlighting major trends such as the explosive growth of coding-related uses and the fast-evolving open vs. closed model landscape. The conversation demystifies common assumptions about AI's present-day impact, and surfaces surprising use cases—including the dominance of roleplay on open-source models.

Key Discussion Points and Insights

1. Industry Headlines & Context

(00:28 – 13:32)

OpenAI's Upcoming GPT-5.2 Release

• Competitive pressure: OpenAI is reportedly fast-tracking GPT-5.2 in response to Google’s Gemini 3 ("Code Red" internally at OpenAI).
• Demand and growth trends: ChatGPT user growth has slowed sharply, while Google-tied AI stocks are surging.
• Market speculation: Polymarket and betting platforms are volatile, reflecting shifting sentiment on which firm leads in AI.

“The stakes are very clearly high for the next iteration of ChatGPT, but the buzz is that the model could live up to the hype.” — NLW (04:11)

OpenAI’s Controversial Product Integrations

• User pushback: Users reported unwelcome suggestions to connect with brands like Target and Spotify, interpreted as ads.
• OpenAI Response: Chief Research Officer Mark Chen admitted these suggestions “felt a lot like advertising” and committed to turning them off and improving control.

“I agree that anything that feels like an ad needs to be handled with care, and we fell short. We've turned off this kind of suggestion.” — Mark Chen, OpenAI (07:58 quote cited)

Major Talent Turmoil at Apple

• Executives departing: Apple's head of AI, head of UX, and general counsel have all departed, with further departures possible.
• Implications: NLW frames this as a major loss of talent for Apple’s AI initiatives, with ongoing instability.

Meta’s Strategic Moves

• Acquisitions: Meta acquires Limitless, an AI wearable startup—interpreted as a talent acquisition for Reality Labs.
• AI News Features: Meta’s chatbot now provides up-to-date news via deals with leading publishers, quietly shifting copyright and licensing dynamics.
• Legal battles: New lawsuits (e.g. Perplexity vs. major news outlets) signal coming courtroom fights over AI-generated content.

2. Study Spotlight: 100 Trillion Token Analysis with OpenRouter & a16z

(13:32 – End)

About the Study

• Data Scope: 100 trillion tokens, focused on developers and power users (not full end-user population).
• Platform: OpenRouter—API gateway for access to 300+ LLMs, serving consumer-facing AI apps and advanced users.

Major Findings

A. The Paradigm Shift: Reasoning Models Dominate

• Token usage:
  • In December 2024, almost no usage was for “reasoning models.”
  • By late 2025, reasoning now accounts for over 50% of all tokens.
• Significance: Marks a fundamental change in AI utility from basic text operations to advanced logical tasks and agentic behaviors.

“The balance between reasoning versus non-reasoning tokens completely shifted… Reasoning Model Token usage went from almost negligible to now over 50% of tokens consumed. OpenRouter calls this a full paradigm shift.” — NLW (15:20)

B. Explosive Growth in AI Coding Use Cases

• Token share: Early 2025: programming 11% of usage → now over 50%.
• Longer prompts: Average prompt size up 4x this year (from ~1,500 to 6,000 tokens).
• Concrete shift: Most token consumption now comes from “Here’s a pile of code, docs, and logs. Now extract the signal.”

“Any accounting of 2025 has to start with the fact that the dominant and most important phenomenon of this year in AI was the rise of AI coding.” — NLW (16:57)

C. Roleplay and Creative Dialogue: The Other Dominant Use Case

• Especially on open-source models, more than 50% of use involves roleplay—from innocent chat to NSFW creative writing.
• On Chinese open source models, programming and technology use cases now surpass roleplay (roleplay now at 33%).

D. Open vs. Closed Models: A Blended Stack

• Open source ascendancy: Open-weight models reached ~33% of usage, plateaued this quarter as closed models advanced.
• Chinese models: From 1% to 30% market share in some weeks.
• Usage distinction:
  • Closed models: high-value workloads.
  • Open models: high-volume workloads.

“If you want a single picture of the modern stack, closed models are for high value workloads and open models are for high volume workloads. And… teams are using both.” — NLW, paraphrasing OpenRouter (18:52)

E. The “Cinderella Glass Slipper” and Model Stickiness

• New models see a burst of experimentation; persistent users create a “foundational cohort.”
• Cohorts that adopt a model early (e.g., Claude 4, Sonnet, Gemini 2.5 Pro) show much lower churn even after many months.

F. Price Insensitivity and Differentiation in AI Models

• Users are willing to pay 10-50x more per token for models that save meaningful time, especially for technical tasks like debugging.
• The market values performance over price.

3. Notable Quotes & Memorable Moments

• Tang Yan (Chain of Thought AI newsletter, cited at 21:55):
  “Anthropic's Claude is used for over 80% of programming and almost zero roleplay. It is the serious work model, while Deep Seek is the Entertainment king with 2/3 roleplay traffic.”
• Tang Yan, on lock-in:
  “A model that’s the first to nail a painful workload creates near permanent lock.”
• Sean Chahan:
  “Role playing in creative writing is 52% of open source usage. While VCs fund productivity, humans are using AI to write fan fiction and debug code. The market gap versus reality gap is hilarious.”
• Brian Catano (meta analysis, 27:11):
  “The AI market is special in its sensitive differentiation. It's easy to switch between providers, but evaluating any model or provider is sensitive… I keep expecting one model to win, but this hasn’t ever really happened.”

Timestamps for Key Segments

• 00:28: Headlines: OpenAI vs Google, stock trends, user metrics
• 04:11: “The stakes are very clearly high…” (market pressure for GPT-5.2)
• 07:58: Mark Chen (OpenAI) acknowledges ad-like integrations
• 09:55: Major departures from Apple’s AI team highlighted
• 11:36: Meta’s acquisition of Limitless and AI wearable strategy
• 13:32: Main Topic: What are people actually using AI for?
• 15:20: Reasoning token usage surpasses 50%
• 16:57: Rise of coding as primary AI use case
• 18:52: Open vs. closed models: differentiated real-world uses
• 21:55: Tang Yan’s analysis on work vs. entertainment model segmentation
• 27:11: Brian Catano on why scaffolds/wrappers (like OpenRouter) thrived

Insights & Takeaways

• Coding has become the dominant AI workload, especially among developers and technically-inclined users, accounting for over half of all usage on OpenRouter’s platform.
• Roleplay and creative dialogue remains a massive segment—over half of open-source model usage centers around “fantasy” conversation, indicating huge and underrated demand for entertainment and social interaction.
• Open-source models—especially from China—are gaining ground fast, now viable for a much wider range of workloads and often viewed as the best choice for high-volume tasks.
• AI model market remains highly fragmented, with no universal “winning” model; developers routinely use both open and closed solutions, seeking specific strengths for specific jobs.
• Users exhibit strong loyalty (“stickiness”) toward early-adopted models that solve genuine pain points—even as new models emerge with incremental improvements.
• Surprisingly, performance trumps price—users and teams are willing to pay outsized premiums for models that grant efficiency or unique capability.

Episode Tone

NLW keeps the episode brisk and punchy, with a subtle thread of skepticism and humor regarding the ways both industry and users are surprising the “experts.” There is an undercurrent of excitement about how fast things are changing, matched with a recognition that hype doesn’t always map to market reality.

For those who missed the episode, this summary should provide not only a rigorous breakdown of what’s driving AI adoption now, but also a sense of where developer and user energy is flowing as we close out 2025.