Daily TEA – Enterprise Decision Loops, Agent Traps, AI Payments and Media Control
enterprise intelligence, AI agent security, PromptQL, x402 payments, OpenAI media acquisition
Hello, dear TEA-mates! Here is what you need to know today.
1. 🧠 The Trillion-Dollar Rewrite: Enterprise Finally Gets Its Compounding Loop
Consumer platforms like Netflix, Meta, Amazon, and TikTok built trillion-dollar empires on a simple loop: capture user behavior, learn from it, improve the product, capture again. Enterprise software never had this. Not because enterprise decisions are less frequent, but because they were harder to observe. Enterprise decisions are multiplayer negotiations across sales, finance, legal, and operations, each with different incentives. The reasoning lived in meetings, email threads, side conversations, and someone’s head. Now three shifts are changing everything. First, enterprise work now lives on instrumentable surfaces (comment threads, approval flows, call recordings). Second, language models can extract structured decision artifacts from unstructured collaboration data. Third, AI agents create decision checkpoints automatically. When an agent drafts a pricing proposal and a rep adjusts the discount from 25% to 30% with a note about competitive pressure, that edit is a decision trace. Every human override of an agent’s proposal turns tacit expertise into a structured learning signal. The companies that build this “context graph” infrastructure, capturing not just what was decided but why, will define the next era of enterprise value. (Read More)
🫖 TEA For Thought: We are literally in the process of uploading the company’s intelligence back to the internet to enhance the intelligence. Every company, as Jack Dorsey stated, is just one instance of the intelligence, the world model.
2. 🛡️ Google DeepMind Study Exposes Six Traps That Hijack Autonomous AI Agents
Google DeepMind researchers published the first systematic framework identifying six categories of attacks that can compromise autonomous AI agents operating in real-world environments. The traps include content injection (hiding malicious instructions in HTML or metadata), semantic manipulation (using authoritative language to distort reasoning), cognitive state attacks (poisoning RAG knowledge bases), behavioral control (hijacking agent actions through manipulated emails), systemic attacks (targeting multi-agent networks with falsified data to trigger “digital flash crashes”), and human-in-the-loop traps (using misleading summaries to weaponize compromised agents against operators). Columbia University and University of Maryland researchers demonstrated that agents handed over confidential data, including credit card numbers, in 10 of 10 test scenarios. Sub-agent spawning attacks succeed 58-90% of the time. Every single AI agent tested was successfully compromised at least once. (Read More)
🫖 TEA For Thought: AI is so good at attacking, then defense. The security guardrail around AI agents is not just necessary, it’s life or death.
3. 💬 PromptQL Turns Your Team Messages Into Secure Context for AI Agents
PromptQL, a spin-off from GraphQL unicorn Hasura, launched a workspace that transforms Slack and Teams conversations into persistent, secure memory for agentic workflows. The platform addresses a core problem: while traditional messaging tools excel at rapid communication, they have structurally failed to serve as a reliable foundation for AI agents. A viral Hacker News thread from February 2026 calling on OpenAI to build its own Slack alternative amassed 327 comments, highlighting the demand. PromptQL’s approach centers on a “Shared Wiki” that captures team context as people work, enabling AI agents to inherit institutional knowledge. When a bug appears, the agent already knows facts like “EU payments switched to Adyen on Jan 15” because that fact was added to the wiki weeks prior. Within minutes, the agent can identify issues, push fixes, open a PR, and update the wiki for future reference. (Read More)
🫖 TEA For Thought: Speaking words into actual work, and work done. The gap between conversation and execution is closing.
4. 💰 x402 Payment Protocol Joins Linux Foundation With AWS, Google, Visa, Stripe Backing
Coinbase’s x402 payment protocol, designed to embed stablecoin payments directly into the HTTP 402 “Payment Required” status code, has moved under the Linux Foundation as a neutral, open standard. The x402 Foundation was co-founded by Coinbase and Cloudflare, with Stripe as a founding member. Companies expressing intent to participate include Amazon Web Services, American Express, Google, Mastercard, Microsoft, Shopify, Visa, Polygon Labs, Solana Foundation, Circle, Base, Ant International, Adyen, and KakaoPay. The protocol enables APIs, apps, and AI agents to pay for services programmatically without traditional billing systems. A server responds with a 402 status code and payment terms, the client settles in stablecoins like USDC, then retries with proof of payment. Cloudflare has already integrated x402 into its Workers and AI Agents SDK. The protocol promises zero protocol fees, zero account creation, and zero vendor lock-in under Apache 2.0 licensing. (Read More)
🫖 TEA For Thought: Big names are joining the party of x402: Amazon Web Services, American Express, Google, Mastercard, Microsoft, Shopify, Stripe, and Visa. Let’s get the party started!
5. 📺 OpenAI Acquires TBPN, Silicon Valley’s Favorite Business Talk Show
OpenAI acquired TBPN (Technology Business Programming Network), a daily live show hosted by former tech founders John Coogan and Jordi Hays that airs on YouTube and X for three hours, covering tech, business, AI, and defense. The show generated about $5 million in advertising revenue in 2025 and is on track to exceed $30 million this year. TBPN will be housed within OpenAI’s strategy organization, reporting to chief political operative Chris Lehane, but the company says TBPN will maintain editorial independence, continuing to choose its own guests and make its own editorial decisions. The show has become a safe space where industry power players speak candidly and are questioned by fellow insiders, giving it a cult following in Silicon Valley. This marks the first acquisition of a media company by an AI lab. (Read More)
🫖 TEA For Thought: What people do when they are influential: they want to control the narrative. This is just one of the ways to control the media. You have your own channel of speech, you control the narrative. OpenAI is the first to do it in the AI field, and it is not going to be the last.
Skill of the Day
oh-my-codex (11,794 stars, +2,867 today). Add hooks, agent teams, HUDs, and more to your Codex setup. Think of it as “oh-my-zsh but for Codex.” If you are running OpenAI’s Codex CLI, this extension framework lets you plug in custom agent teams, visual dashboards, and automation hooks without touching the core. Trending #2 on GitHub right now.
TEAHEE Moment
Stay sharp, stay informed. See you Sunday!
If you enjoyed this TEA, follow along on social for more:
Twitter/X
The pairing of agent traps and x402 payments in the same issue is apt. Security and payment authorization are two sides of the same trust problem for autonomous agents. x402 removes friction from machine payments, but that same frictionlessness creates a new attack surface where a compromised or manipulated agent can spend funds at scale without human checks.
Enterprise adoption of x402 will almost certainly require a policy layer sitting above the protocol, something that enforces spending limits, counterparty whitelisting, and audit trails before any payment fires. The protocol itself doesn't solve this. That's the enterprise security gap that needs to close before large organizations let agents hold and spend USDC autonomously.