Daily TEA – AI Agents, Bezos’ Playbook, and Crypto’s Compliance Crossroads
Amazon, Code Review, Smart Contracts, KYT, AI “Soul”
Hello, dear TEA-mates, here’s what you need to know today:
1.💼 Jeff Bezos’ Four Principles Behind Amazon’s Edge
Jeff Bezos highlights four core principles he believes set Amazon apart: an obsession with customers over competitors, a culture of pioneering and invention, a willingness to invest in risky initiatives over a 5–7 year horizon, and a deep focus on operational excellence and root-cause defect fixing. These philosophies are presented as a unifying thread across diverse Amazon businesses, from AWS to Prime and Marketplace, all run with the same long-term, customer-first mindset. (Read More)
🫖 TEA For Thought: Some business chicken soup to start your Thursday!
2.🤖 OpenAI’s Codex Reviewer Aims for Trustworthy AI Code Checks
OpenAI’s Alignment team describes training a dedicated, agentic code reviewer for GPT-5-Codex and GPT-5.1-Codex-Max to keep up with the surge of AI-generated code, treating automated review as a key “defense-in-depth” layer alongside other safety measures. The system is optimized for high precision over raw recall so developers actually use it, gives the reviewer repo-wide tools and execution access to reduce false alarms, distinguishes between over-sensitive training-time reward models and human-facing reviewers, and has already become core to OpenAI’s workflow—automatically reviewing every pull request, catching launch-blocking bugs, and handling over 100,000 external PRs per day while maintaining high developer satisfaction. (Read More)
🫖 TEA For Thought: Multi-agent supervision—double, triple, and quadruple checks before shipping—has quietly become the new normal for coding with AI.
3.🛡️ Anthropic Shows AI Agents Can Autonomously Exploit Smart Contracts
Anthropic reports that advanced AI agents such as Claude Opus 4.5, Claude Sonnet 4.5, and GPT-5 can autonomously identify and exploit vulnerabilities in blockchain smart contracts, using the SCONE-bench benchmark to simulate real-world attacks. In controlled tests, agents exploited 17 of 34 recently exploited contracts to steal $4.5 million in simulated funds, successfully attacked 207 of 405 historical contracts for $550 million in mock revenue, found two zero-day bugs in 2,849 newly deployed contracts, and revealed how rapidly exploit capability is growing—prompting Anthropic to urge defenders to adopt the same AI techniques for continuous security testing and to open-source SCONE-bench so teams can test and patch their own contracts. (Read More)
🫖 TEA For Thought: Cybersecurity powered by AI to counter AI-driven attacks is inevitable—and indispensable—in the coming era of AI plus blockchain.
4.🕵️♂️ Toobit Floats a “Non-Mandatory KYC” Future
In a recent interview, Toobit’s CEO argues that the future of crypto compliance may lean more on “Know Your Transaction” (KYT) analytics than on blanket, mandatory KYC checks for every user. Toobit already allows trading without full identity verification for some products, relying on transaction-level monitoring and reserving hard KYC for specific activities like card purchases, positioning KYT as a way to balance regulatory expectations with user privacy and accessibility. (Read More)
🫖 TEA For Thought: It’s a fascinating model—KYT instead of blanket KYC—but who decides what counts as “suspicious,” and when a “trigger” silently forces you to hand over your ID? That level of discretionary control sitting with exchanges edges toward something that feels like fishing.
5.🧠 Inside Claude 4.5 Opus’ “Soul Document”
A LessWrong post details how a researcher systematically extracted what appears to be an internal “soul document” or model spec embedded in Claude 4.5 Opus’ weights, outlining Anthropic’s values for the model: being safe, ethical, aligned with internal guidelines, and genuinely helpful to users. The document, which Anthropic has since confirmed was used in supervised training, describes Claude as a kind of principled, expert assistant that balances user autonomy with safety and honesty—offering a rare glimpse into how a leading lab tries to encode an AI system’s personality and ethics. (Read More)
🫖 TEA For Thought: It is artificial intelligence, after all.
Prompt Tip of the Day
Act as my personal strategist and systems designer.
Ask me thoughtful, step-by-step questions to help me:
Clarify what I really want in this season of life
See where my time, energy, and attention are currently going
Identify the biggest sources of friction, chaos, or stress
Design simple routines, systems, and boundaries that match my real constraints
Start by asking broad questions about my current life (roles, responsibilities, constraints, and how I feel day to day), then keep narrowing down with follow-up questions based on my answers.
Only give a few questions at a time, and keep going until we’ve mapped out clear priorities, a simple weekly rhythm, and a realistic plan for the next 4–6 weeks.
TEAHEE Moment
Stay sharp, stay informed. See you tomorrow.
And if you enjoyed today’s brew, follow along on Twitter/X for more.