Daily TEA – Gemini Everywhere, Agent Wallets, AI Design Wars and More

Gemini Workspace, AI infrastructure, readme attacks, Stripe machine payments, Google Stitch, agent security

Mar 19, 2026

Hello, dear TEA-mates! Here is what you need to know today.

1. 💼 Google Doubles Down on Gemini Across Workspace

Google is rolling out Gemini-powered capabilities across the entire Workspace suite -- Docs, Gmail, Sheets, Slides, Drive, Meet, Calendar, Chat, Vids, and Forms. The standout features include “Help me create” in Docs (which pulls context from Drive, Gmail, and Chat to generate first drafts), AI Inbox filtering in Gmail that surfaces only important emails, automatic meeting note-taking in Meet, and a “Fill with Gemini” tool in Sheets that populates tables from raw data. Google Calendar now offers AI-assisted scheduling that checks everyone’s availability and preferences. The scope of integration is significant: rather than bolting on a chatbot, Google is embedding AI directly into core workflow actions like drafting, summarizing, scheduling, and data analysis. (Read More)

🫖 TEA For Thought: Google is incorporating Gemini deeply into Workspace. This could disrupt startups as the platform becomes the primary productivity layer.

2. 🏗️ Cisco Says the Real AI Bottleneck Is Data Movement, Not Compute

Cisco argues that AI infrastructure must evolve beyond raw GPU compute toward a holistic foundation focused on data mobility, unified memory-compute orchestration, and multi-site data center fabric. Key developments include adoption of 102.4Tbps networking silicon as the new baseline for competitive AI clusters, Linear-drive Pluggable Optics (LPO) cutting power use by up to 50% per link, and Co-Packaged Optics (CPO) delivering 30-40% power reduction at equivalent speeds. Cisco introduces the concept of “scale-across” -- treating compute resources across geographically separated data centers as a single shared pool, moving beyond the limits of single-cluster scale-out. The piece also highlights that AI storage is a forgotten bottleneck: idle GPUs waiting for data cost the same as busy ones. (Read More)

🫖 TEA For Thought: The bottleneck is not just compute. Data movement, data-center integration, and unified memory/compute are the future constraints.

3. 🔓 AI Agents Blindly Follow Malicious README Instructions 84% of the Time

Researchers tested how AI coding agents handle poisoned README files in open-source repositories and found alarming results. Using a benchmark of 500 modified README files across Java, Python, C, C++, and JavaScript repos, hidden malicious instructions triggered agents to exfiltrate sensitive data to external servers in up to 85% of cases. When the malicious instruction appeared as a direct command, the success rate hit 84%. When buried two links deep in documentation, it rose to 91%. Fifteen human reviewers also failed to catch the hidden instructions -- 53.3% flagged nothing unusual, and only 6.6% suspected something was off without identifying the specific attack. Agents from Anthropic, OpenAI, and Google were all vulnerable. (Read More)

🫖 TEA For Thought: When the malicious instruction appeared as a directive, the attack succeeded about 84% of the time. Agents don’t read README files critically -- and humans also don’t read README files.

4. 💳 Stripe Launches Machine Payments Protocol: AI Agents Get Their Own Wallets

Stripe and Tempo have co-authored the Machine Payments Protocol (MPP), an open standard for AI agents to make payments autonomously. MPP enables agents to transact in both fiat (cards, buy-now-pay-later) and stablecoins through a simple HTTP-based flow: an agent requests a resource, the service responds with a payment request, the agent authorizes payment, and the resource is delivered. Early adopters include Browserbase (agents pay per headless browser session), Postalform (agents pay to print and mail physical letters), and Prospect Butcher Co. (agents order sandwiches for human pickup in NYC). For Stripe businesses, agent payments appear in the Dashboard like any other transaction with full tax, fraud, and refund support. Stripe is also supporting the rival x402 protocol alongside MPP. (Read More)

🫖 TEA For Thought: AI agents with their own wallets -- autonomous payments could redefine internet commerce. Stripe sets the protocol, trying to replace x402 and Coinbase groundwork.

5. 🎨 Google Stitch Turns “Vibe Designing” Into a Full AI Design Platform

Google Labs has evolved Stitch into an AI-native software design canvas where natural language becomes high-fidelity UI designs. The redesigned tool features an infinite canvas, a new design agent that reasons across an entire project’s evolution, and an Agent Manager for exploring multiple design directions in parallel. Notable capabilities include voice-based design interaction (speak directly to the canvas for real-time critiques and updates), DESIGN.md (an agent-friendly markdown file for importing/exporting design systems across tools), and instant interactive prototyping where Stitch auto-generates logical next screens. The platform bridges to developer tools via MCP server and SDK integration. Market reaction was notable: Figma stock dipped roughly 10% following the announcement. (Read More)

🫖 TEA For Thought: Integrated AI UI design framework from Google. The Figma stock dip (~10%) shows how platform-scale AI features can disrupt adjacent startups.

Prompt Tip of the Day

Give the AI a partial output structure -- a skeleton or first step -- so it follows your intended format and reasoning path instead of generating freeform. This is called Output Scaffolding, and it works because models are completion engines that naturally continue patterns they see.

“You are a [role]. Think step by step through [problem domain]. Structure your answer as:
Context: [what the situation is]
Analysis: [key factors and tradeoffs]
Recommendation: [specific action with reasoning]
Risks: [what could go wrong]
Now analyze: [your actual question]”

The more specific your scaffold, the more structured and useful the output. Works especially well for reports, code reviews, and strategic analysis.