Daily TEA – Intelligence Flows, Local Flavor, DePIN Dreams & Dynamic AI

Perplexity, Adobe trends, PancakeSwap, DePIN, ChatGPT UI

Dec 18, 2025

Hello, dear TEA-mates, here’s what you need to know today.

1.🤝 Perplexity & NotebookLM Win With “Intelligence Flow Architecture”

Product designer Adrian Levy argues that Perplexity and Google’s NotebookLM succeed not because they use superior AI models, but because they architect the flow of cognitive work between humans and AI differently. Both products run on common building blocks like large language models, retrieval-augmented generation and search APIs, yet feel distinct because they assign discovery, synthesis and orchestration to AI while leaving goal-setting and judgment to humans. Perplexity’s design has AI autonomously search, synthesize and cite sources so users simply direct and evaluate, whereas NotebookLM focuses AI on understanding and structuring a user’s own documents into a knowledge graph that can power multiple formats. Levy calls this “Intelligence Flow Architecture” and outlines principles such as designing backward from autonomous execution, embedding intelligence in specific layers, defining clear human/AI boundaries and optimizing cognitive distribution between human strengths and AI strengths. (Read More)

🫖 TEA For Thought: This approach treats AI as a true collaborative partner with clear boundaries, optimizing for each side’s strengths: AI excels at rapid processing and pattern recognition, while humans provide direction and judgment.

2.🌍 Adobe’s 2026 Creative Trends Highlight Emotion, Playfulness and Local Flavor

Adobe’s 2026 Creative Trends forecast identifies four major themes: “All the feels,” “Connectioneering,” “Surreal Silliness” and “Local flavours.” “All the feels” emphasizes multi-sensory branding experiences, encouraging creatives to craft visuals that evoke texture, sound and even taste, including richer prompt design for generative AI. “Connectioneering” focuses on emotionally resonant storytelling where shared joy, inspiration and affection help audiences feel the message rather than notice the branding. “Surreal Silliness” spotlights playful, logic-bending work enabled by generative AI tools like Adobe Firefly, with research suggesting fun, playful brands are better at building future demand. Finally, “Local flavours” reflects Gen Z’s move toward “appstinence” and real-world presence, pushing brands to partner with local creators and spotlight regional craftsmanship in authentic, unfiltered ways. (Read More)

🫖 TEA For Thought: The emphasis on local flavor genuinely surprises me; in the age of AI, the authenticity of real life seems to outperform anything purely digital.

3.🎯 PancakeSwap-Backed Probable Launches Zero-Fee Prediction Markets on BNB Chain

Probable, a zero-fee prediction markets platform incubated by PancakeSwap and supported by YZi Labs, is preparing to launch exclusively on BNB Chain. The platform plans to offer fast, transparent markets across crypto, global events, sports and regional topics, with users able to deposit any token that is automatically converted into USDT on BNB Chain for seamless onboarding. Probable emphasizes a simple, speed-focused interface and on-chain transparency, aiming to avoid the cluttered designs that often complicate traditional prediction markets. Security will be powered by UMA’s Optimistic Oracle, which uses dispute-driven validation to provide tamper-resistant event verification and economically guaranteed outcomes across diverse markets. The launch comes amid a broader boom in prediction markets since 2024, with platforms like Kalshi and Polymarket driving billions in monthly volumes and attracting attention from major crypto firms, even as regulatory uncertainty around event contracts and gambling persists. (Read More)

🫖 TEA For Thought: This looks like yet another way for Binance to edge further into prediction markets and quietly vacuum up people’s wallets.

4.🧮 Inference Economics: Reserved Compute vs Inference APIs

Data Gravity’s “Inference Economics 101” explains how AI infrastructure is splitting into two durable markets: reserved/hourly GPU platforms focused on predictability and control, and inference APIs optimized for utilization efficiency and cost abstraction. Reserved compute platforms sell guaranteed access to GPUs, stable performance and full stack control, yielding high utilization per customer, larger contracts and lower churn, but growth that scales linearly with fleet expansion. Inference APIs, by contrast, let customers buy tokens or requests rather than hardware, aggregating bursty workloads from many users to achieve higher platform-level utilization and better effective costs, at the expense of some per-request determinism. The piece argues that utilization — how busy GPUs are — is the dominant driver of inference economics, with performance gains from faster tokens-per-second often overshadowed by idle capacity, especially for user-facing, bursty workloads. Because many performance techniques are open source and diffuse quickly, the article concludes that durable advantage will accrue to platforms that master aggregation, abstraction and control planes rather than those that merely push raw execution speed. (Read More)

🫖 TEA For Thought: This further reinforces the view that DePIN, where AI compute is decentralized and users pay in crypto to access other users’ compute, is the most compelling future.

5.🌀 ChatGPT Pushes Beyond Text Toward Dynamic, Multimodal Experiences

OpenAI’s Fidji Simo describes how ChatGPT is evolving from a text-centric chatbot into a more dynamic, multimodal and app-connected experience. A new image generation model now lives inside a dedicated image-focused entrypoint that works like a creative studio, with improved viewing and editing tools, faster performance and better adherence to detailed instructions while preserving elements like lighting and composition. ChatGPT is also bringing richer visuals into answers — including photos, side-by-side specs and tappable entities for people, places and products — so users can explore context inline rather than issuing repeated follow-up prompts or switching tools. For everyday utilities like unit conversions or sports scores, the product is rolling out concise visual responses and is revamping writing with inline “canvas”-style blocks, better export options and smoother flows into email, messaging and document apps. A growing ecosystem of apps from partners such as Booking.com, Canva, Figma, Spotify, Zillow, Adobe, Airtable, Replit and Salesforce is turning ChatGPT into a generative UI that surfaces the right tool at the right time, aiming to narrow the gap between what users imagine and what they can create. (Read More)

🫖 TEA For Thought: Expect many more products like this—traditional chat UIs are too constrained, and the future will lean heavily into voice control plus rich image and video interactions.