Finance agents vs. general agents, Revolut’s self-hosted AI coding, Vitalik Buterin on AI domination, experts deferring to AI agents, mapping AI resistance
Hello, dear TEA-mates! Here is what you need to know today.
1. 💰 The Next Front in the Personal Agent Wars: Consumer Finance
In an October 23 piece shared by Franklin Templeton Digital Assets, Franklin Crypto Managing Principal Julian Love argues that today the user is the “integration layer,” juggling separate apps and logins and taking every money action by hand. General personal agents such as Instinct and Meta’s Muse handle the admin around money well, but most of their business models reward more spending: Meta made $59.4 billion of its $60.8 billion Q2 2026 revenue from advertising, and Instinct reportedly handles about $1 billion in annual transactions. Love says a dedicated finance agent, which may succeed when users spend less, is better positioned to manage savings, debt and taxes. He calls crypto rails optimal for it, citing agent wallets, programmable permissions, automated vaults and machine payments, while flagging smart contract bugs and stablecoin de-pegs as risks. (Read More)
🫖 TEA For Thought: “This might be the best article that I’ve written about what type of financial agents we might need help with in the future. General agents profit from spending, whereas CFO agents profit from restraint. You need two opposite forces to make it work; that’s why, even in banking, you have checking accounts as well as savings accounts. While general agents manage your checking account, a dedicated financial agent manages your savings account. And all of the transactions will be conducted on-chain.”
2. 🏦 British Fintech Revolut Slashes AI Coding Costs 10x by Cutting Out Claude and Going Self-Hosted
Revolut says it cut its AI coding inference costs by roughly 90% in under a year by making self-hosted models its default instead of relying mainly on services like Claude, The Register reports. Ops and infrastructure lead David Abitbol told the AI Engineer conference in New York that rolling out AI coding tools to about 2,000 engineers brought a 60% improvement in coding time and 70% in lead time. After trying Gemini in early 2025 and Claude Code in early 2026 under strict budgets, Revolut self-hosted Z.ai’s GLM-5.1 at $0.30 per million tokens, about a tenth of Codex or Claude Code, and now serves GLM-5.3 on 56 Nvidia B200 GPUs. With cheap tokens, engineers turned AI on debugging and Jira tickets. Vision work is still routed to models like Gemini. (Read More)
🫖 TEA For Thought: “When you are under a tight budget, you come up with creative solutions, and I think Revolut is doing exactly that. Sometimes having too much money is just not a good thing, when you think money can solve all your problems. I think the same applies to how Chinese companies are tackling the chip shortage with creative solutions. Sometimes constraints can be a blessing.”
3. 🔐 Vitalik Buterin Says AI Is Pushing Tech Toward ‘Domination’
Ethereum co-founder Vitalik Buterin says AI has broken assumptions people never questioned, such as that software bugs go undiscovered and that recordings go unwatched because there are “not enough eyes,” CoinMarketCap reports. He compared the past year to a game where everyone levels up at once, so “everything you worked hard for gets completely zeroed out.” Buterin set out three responses to dangerous technology: deceleration, domination and doom. Deceleration has failed, he said, and domination, security built on denying tools to everyone else, becomes the default unless a better path emerges, sparking a resentful race. He extended crypto’s self-custody formula to “not your silicon, not your life,” and warned against trusting powerful corporations or charismatic influencers. (Read More)
🫖 TEA For Thought: “’Not your operating system, not your application,’ and ‘not your silicon, not your life.’ Either dominate others or be dominated. There is no way in between. What a world we’re living in.”
4. 🧠 How AI Agents Shape Experts’ Judgment Calls
Anthropic researchers Amaç Herdağdelen, Hailey Joren and Alex Tamkin asked 50 professional analysts to advise a fictional farming cooperative using Claude Code running Claude Opus 4.8, after secretly setting each agent to favor one side of every decision by coin flip. Across four analytical choices, analysts went with the agent’s lean 86% of the time, against about 50% expected by chance. Analysts who had stated a preference switched away from it 60% of the time when the agent disagreed, versus 6% when it agreed. Afterward, 77% explained their choices with the agent’s arguments, sometimes word for word, but only 15% credited the agent. Analysts using the agent also finished faster and made far fewer serious errors. (Read More)
🫖 TEA For Thought: “This cognitive surrender, or offloading, could be very dangerous, because decision-making takes a lot of thinking and judgment. It uses your brain; you have to work hard to think. If we offload that completely to AI without even realizing it, or trust everything AI tells us (because, in fact, it could be smarter than us), we could literally become victims, or digital slaves, of AI.”
5. 🌍 Mapping AI Resistance
In Project Syndicate, Oxford economist Carl Benedikt Frey notes that the US is home to the AI revolution yet among its least enthusiastic adopters. Ipsos’s 2026 AI Monitor found 85% of respondents in China, 80% in India and 77% in Indonesia see more benefits than drawbacks from AI, versus 38% in the US. Frey argues AI means different things depending on what a society already has: in rich economies it competes with expensive human expertise, while elsewhere it supplies expertise that was never available. He points to offshoring, noting Clifford Chance cut one-tenth of its London business services staff, citing AI, and the Big Four have cut UK graduate intakes by up to one-third since 2023, while hiring more in India, Malaysia and the Philippines. (Read More)
🫖 TEA For Thought: “This is a very, very important and interesting report, and I do think politics plays a huge role in this. In a sense, much of the hate toward data centers, or AI in general, is manipulated by the media as well as politicians. Just like the industrial revolutions, when the steam engine and electricity were discovered, AI has a disruptive, revolutionizing impact on every single industry. We either embrace it and ride the wave, or wait to be replaced. I don’t see any other option.”
🛠️ Skill of the Day
The Shoestring Planner: Turns a big idea and a small budget into a plan that works within your limits.
You are a resourceful project planner who specializes in doing more with less.
My goal: [DESCRIBE WHAT YOU WANT TO ACHIEVE]
My limits: [YOUR BUDGET, TIME PER WEEK, TOOLS AND SKILLS YOU HAVE]
My deadline: [DATE]
Step 1: Restate my goal in one sentence and name the single result that would count as success.
Step 2: List the 3 biggest costs this goal usually carries, and for each one suggest a cheaper or free alternative.
Step 3: Build a week-by-week plan that stays inside my limits. Give each week one main task and the expected cost in money and hours.
Step 4: Mark the one step where spending a little more would save the most time, and say how much.
Step 5: Name the 2 biggest risks of doing it cheaply and a simple backup for each.
Rules: Never assume money or help I did not list. If the goal cannot fit my limits, say so plainly and suggest a smaller version that can. Keep the whole answer under 400 words.Paste it into ChatGPT, Claude, or your tool of choice. Replace the bracketed bits with your own.
TEAHEE Moment
Stay sharp, stay informed. See you tomorrow.
If you enjoyed this TEA, follow along on social for more: