Daily TEA – On-Device AI, Data Markets, and the New Software Order
SnapGen, Cloudflare x Human Native, data centers, BRICS digital currency, Software 2.0
Hello, dear TEA-mates—here’s what you need to know today.
1.🎨 Snap’s SnapGen++ Brings High-Res On-Device Image Generation to iPhone
Snap has introduced SnapGen++, a compact diffusion transformer model that generates 1024×1024 pixel images directly on iPhones in about 1.8 seconds, eliminating the need for cloud processing. By running entirely on-device and leveraging Apple’s neural and GPU hardware, SnapGen++ delivers image quality comparable to much larger models while improving privacy and reducing latency. Read More (https://the-decoder.com/snaps-snapgen-generates-high-resolution-ai-images-on-iphone-in-under-two-seconds/)
🫖 TEA For Thought: Models this small will only get more popular—private, faster, and cheaper.
2.☁️ Cloudflare Acquires Human Native to Build an AI Data Marketplace
Cloudflare is acquiring Human Native, a UK-based AI data marketplace that connects content creators and publishers with AI developers seeking high-quality training and inference data. The deal aims to create a new economic model for the internet by giving creators more control over how their content is used and enabling transparent, paid transactions when AI companies train models on their work. Read More (https://www.cnbc.com/2026/01/15/cloudflare-ai-human-native-acquisition.html)
🫖 TEA For Thought: This deal feels genuinely revolutionary—Human Native is becoming one of the leading data marketplaces in the AI era, and it raises big questions about how creators will get paid for their data; blockchain still seems like the most efficient way to scale that.
3.🧰 AI Data Centers Face a Shortage of Electricians and Plumbers
The rapid expansion of AI data centers in the US is running into a serious bottleneck: there are not enough skilled electricians, plumbers, and HVAC technicians to build and maintain these highly specialized facilities. Because AI data centers require far more complex power and liquid cooling systems than typical buildings, projects are being slowed by labor constraints even as wages and overtime opportunities rise for tradespeople qualified to handle the work. Read More (https://www.wired.com/story/why-there-arent-enough-electricians-and-plumbers-to-build-ai-data-centers/)
🫖 TEA For Thought: For now—and likely for a long time—blue collar really is the new white collar, because without data centers, there is no AI.
4.🌐 India Moves to Link BRICS Digital Currencies
India’s central bank has proposed linking the central bank digital currencies of BRICS members to facilitate cross-border trade and tourism, and has asked to place the plan on the agenda of this year’s BRICS summit it will host. The initiative would connect pilots like India’s e-rupee and China’s digital yuan, with discussions expected to focus on interoperability, governance, and how to handle trade imbalances, even as officials insist the move is not explicitly aimed at de-dollarisation. Read More (https://coinmarketcap.com/academy/article/indias-rbi-proposes-brics-digital-currency-network)
🫖 TEA For Thought: This marks another step toward de-dollarization within the BRICS community.
5.💾 Rethinking Software as AI Agents Reshape the Stack
In a new essay, Fabricated Knowledge argues that AI agents and their context windows are becoming the “fast memory” of computing, while software evolves into persistent infrastructure more akin to NAND storage—a single source of truth for data, state, and APIs. The piece contends that human-facing horizontal software and classic SaaS differentiation (better UI, smoother workflows) will lose value as AI agents take over consumption, making persistent, API-first systems and data layers the real winners. Read More (https://www.fabricatedknowledge.com/p/the-death-of-software-20-a-better)
🫖 TEA For Thought: When anyone can spin up an app in minutes to solve a problem, paying for traditional SaaS starts looking like a bad deal.
Prompt Tip of the Day: Ask Me Questions
I’ll call it “Ask Me Questions” (AMQ). It’s like this. Put all the below into 1 prompt:
AI’s role: “You’re an experienced front-end developer...”
What you need: “Today you need to implement: X, Y, Z..”
End with: “Before I change you to Agent mode so you can actually implement, do you have any questions for me?”
Then submit the prompt in “Ask” mode.
The model will ask you insightful, clarifying questions that cover the inputs you didn’t provide yet. It is so much better than just hoping for a successful one-shot.
You can repeat this as many times as needed until you’re convinced the model is well-positioned to succeed.
Bonus tip: for cost-optimization, you can run the questions through a more expensive model, then ask it to make a plan, then defer the implementation to a cheaper model.
TEAHEE Moment
Stay sharp, stay informed. See you tomorrow.
If you enjoyed today’s TEA, follow along for more real-time sips of news and threads on X: Twitter/X.