Things I wish I could do with AI (that is still surprisingly hard) 🧵

The issue with Wise, especially when dealing with USD, is that they are not the actual bank handling the inbound/outbound funds: the real bank behind the scenes is Community Federal Savings Bank (CFSB). This is what "CFSB" looks like on Street View. You think you're interacting with Wise, but you're not - the full compliance and payment rails is running through this mom-and-pop bank. There is a huge mismatch in the size of this bank and their capability to deal with it and what a client like Wise puts through their system. More often that not, any significant transfer will likely get flagged for manual processing. Wise is slightly better when dealing with GBP or EUR as they do have in-house entities with relevant licenses, so they can have better automated processes. The dirty secret of a lot of fintech is so much of it is a shiny API/UI wrappers around very, very legacy infrastructure.

Q. What do you ask when your team is missing one legendary researcher to build foundational LLMs? A. Il y a Sutskever ? (needs knowledge of French and English)

What in the fresh pseudoscience is a “greenspector” certificate? Basically seems like run Playwright journeys and then apply some magic multiplier based on how long tests run. greenspector.com/en/home/

Welp, that happened faster than I predicted. Thought it would be end of 2027, then early 2027, but agentic traffic growing so fast that bots have now passed human traffic online for the first time in the Internet's history. radar.cloudflare.com/traffic…

Love this collab with Blockscout to demystify how FHE-powered transactions work on Ethereum - a great tool for anyone building privacy applications on @zama!

The EU used to be a leader in privacy-preserving digital ID for a very long time (2017 to 2024 or so), but has unfortunately since ceded ground to groups that want zk/decentralised only “in theory”. They’ve got a maximum of 2-3 year more years they can keep coasting on their previous reputation as a supporter of such privacy tech. But with the way things are heading, they are simply not serious about any of their zk/decentralised work, and the factions wanting to push through centralised, state-controlled tech are growing stronger.

Many such instances where confidentiality is important 🤭

We love you too @claudeai

One of the wildest design quirks you’ll find in Indian e-commerce (especially on the Indian Railways booking site) is that the UI forces the user to choose the payment processor. At a glance, it looks like bad UX. Why make the user choose between PayU, Razorpay, or NTT Data? If the goal is resiliency, shouldn't automated "smart routing" just dynamically handle failovers behind the scenes? But if you scratch beneath the surface, it’s actually a masterclass in market-specific engineering. It's actually quite clever: 1. Mind-boggling scale: About 50% of ALL online payments globally happen in India. By comparison, US's share of online payments is ~1-2%, Europe is about ~3-5%. India's real-time transaction volume is huge: 100 billion transactions yearly! That number is greater than the transaction volumes in the US, UK, Canada, France, and Germany combined...and then multiplied by eleven. (Where India's share is smaller is transaction value. If you want an insight into the future of micropayments, look here. Because the sheer infrastructure strain is immense, no single traditional payment processor can guarantee 100% uptime.) 2. Power-user overrides: If a Stripe payment fails on a regular e-commerce site, waiting 10-15 minutes while the system figures out what happened is annoying but fine. But on Indian Railways, high-demand tickets sell out in seconds. You're competing with thousands of people for a fixed number of seats releasing at a specific moment. Those 10-15 minutes could be the difference between getting the ticket or not. So, counter-intuitively, users want to be power users. They'd rather immediately retry with a different processor, than wait for automated fallback logic to catch up. 3. Anti-monopoly by design: On publicly-funded services such as this, locking into a single payment processor would hand one company a dominant position over an enormous share of transaction volume...and the processing fees that come with it. The multi-processor setup avoids creating that winner-takes-all dynamic. In fact, the order in which these payment processors are shown is also randomised on every page load, so that the order on the screen itself doesn't disproportionately drive volume to a single processor. 4. Transaction costs aren't always absorbed by the merchant: Unlike the West, where merchants quietly absorb a 2-3% card processing fee, in India the processing fees are often directly passed to the user as a "convenience fee." Different processors negotiate different rates with specific banks, and some pass on the savings to the users, which creates a live competitive market. Users actively pick the processor with the lowest fee for their transaction based, or cashback provided based on special promotions. It's a local version of pointsmaxxing. There's enormous richness and complexity in India's payments ecosystem, and more broadly across the global south. It is a fascinating look at high-velocity financial infrastructure built under real constraints: scale, competition, scarcity, that Western fintechs largely never had to contend with. But that same ecosystem is also ripe for disruption. Moving payments and finance onchain will be a generational shift in speed and cost, but only if the privacy problem gets solved. No national or local payments infrastructure can run without that, as financial data is some of the most sensitive data there is.

The THORchain $10.8M is gone - let's protect the next protocol or wallet. Here is a short list of ECDSA TSS protocols and libraries that should not be in production right now. The list exists. The deprecations are documented publicly, please follow them:

Hard to believe my first song ever on Spotify was Shania Twain 😂

As crypto matures into financial infrastructure, moats will be defined by security posture, compliance enablement and confidentiality. We keep treating privacy as a silo - a category or a chain you go to. Security stopped being a product the moment it became table stakes. Privacy will have the same arc, and those that embrace this sooner than later will own the rails everyone else ends up renting.

The team has been working hard on this one. It's not a refactor or a rewrite, it is a totally new product that is helping multiple teams deliver @zama protocol integrations in record times.

It’s 2026, and it’s still not possible to provide Gemini with custom instructions/customisation. In the long line evidence of “Google does not understand how to build consumer apps” 🫠

This has been an exciting project to work on in the past few months, to reimagine what building on Zama Protocol looks like. The core design principle our team had in mind: how close can we get the developer experience to, say, something like ERC-20? Not “how do we expose FHE to developers” but “how do we make FHE invisible to them.” Confidentiality on blockchains shouldn’t require a developer to understand the internals of advanced cryptography and mathematics – the same way they don’t need to understand the full HTTPS stack and its library of associated specs to ship a secure web app. That complexity exists; it just shouldn’t be your problem. One north star metric I’ve always used when building products: can someone build a *meaningful* app in under 30 minutes, starting from scratch with no prior background? That benchmark is getting even more interesting with widespread access to AI coding tools. We’ll be releasing official Zama skills for the SDK shortly, enabling anyone to build meaningful, secure, production-ready apps with confidentiality baked in. Faster than 30 minutes. A lot faster. Clear-text in, clear-text out. That’s the principle. Everything else is our problem @zama to solve, not yours.

SHIELDED PREVIEW → Fastest privacy protocol to reach $100M TVS → 45% of circulating $ZAMA staked → #1 app on Ethereum during the Auction Public release tomorrow. Stand by! zama.org/shielded