Grigore Rosu

FastSet is the most significant milestone of my technical and academic career since creating the K Framework in 2003. I didn’t set out to build another blockchain, we have enough fragmentation, brittle bridges, and total ordering slowing Web3 down. FastSet is something else entirely: lightning-fast, decentralized, and verifiable by design. It solves the blockchain trilemma and opens new possibilities for AI, DeFi, and beyond. The future of Web3 isn’t blockchain. It’s verifiable settlement. Try it for yourself and ask me anything.

@azaztrader01 Timestamps on claims. Unlike a blockchain, where the global order decides the timestamps as well on the transactions, we had to find another way. It is actually very clever and the right thing to do.

@RosuGrigore Sounds like solid progress on the backend. What's been the most interesting technical challenge your team has tackled recently?

Pi Squared Weekly Update Infrastructure maturation and production readiness. Teams focused on scalability, reliability, and supporting real-world use cases as adoption pathways expand. Here is what is new ↓

AI agents need super-fast super-cheap payments. Neither blockchains nor traditional money rails can sustain what is coming. Fast (@PiSquared) next gen infra for crypto is here, from Marc's @pmarca 's Alma Mater (UIUC).

@alexabelonix @PiSquared Nice!

@RosuGrigore @PiSquared I had the agent cop a pair of Bluetooth earbuds, funded with USDC so checkout was instant, tracking link popped up in under a minute and delivery was smooth

Your AI agent can now buy things on Amazon while you're on the go. No card. No login. No Amazon account needed. Fund with USDC → tell it what you want → it orders it and sends you a tracking link

@alexabelonix @PiSquared if you have an agent, buy something on amazon and let us know how it works

@PiSquared super solid work.

@SFERAeye @PiSquared Try it and let us know.

@PiSquared wait this is kinda crazy if it actually works smoothly this changes how people interact with online shopping

@LindsayCasale @rv_inc @PiSquared Thank you @LindsayCasale !

@RosuGrigore @rv_inc @PiSquared Fast is the future of payments for agents. Congrats @RosuGrigore and team!

30 years of research, two companies @rv_inc @PiSquared Formal methods, semantics, logic, concurrency, cryptography - all converging into one thing: agents that can pay anyone at the speed of the internet No cards. No logins. No friction Just getting started

@Txn_tochi @rv_inc you need an AI agent. send it the link and it will figure it out by itself. you just play along

@RosuGrigore @rv_inc How to use ser

@Mike_nike7 @PiSquared also cards and USD in general, soon. very soon

@PiSquared this is the use case that actually moves normies. not defi, not yield, just... buy stuff with crypto and it works

@pipsandbills @PiSquared surprise :)

@PiSquared so Fast is launched on mainnet already??

@Akashsoneium @rv_inc Thank you. Try it, works with real money and real products.

@RosuGrigore @rv_inc Looking amazing

@andyyy Not insane, Andy, it is totally obvious. At this point, it is simply survival for CC issuers to take action. What they don't know though, and most crypto people don't know either, is that blockchains are not the solution either: too slow, too expensive.

Mastercard representative just said on a panel at EthCC that cards aren’t great for agentic commerce because of the demand for micropayments & they are seriously considering the massive change in demand profile. Kinda insane tbh

@hosseeb When it happens, in a few months, blockchains are not ready for the required transaction volume. No existing payment system is.

Agentic payments will be a huge trend. But it's not here yet. I had the most bearish take in this article, and I want to explain why, because crypto has a bad habit of getting high on its own supply and giving unrealistic timelines about technology. Take the mouse. This is what the first computer mouse looked like. Imagine seeing this thing and peering into the future--no more terminal interfaces, now GUIs would onboard billions onto personal computers. You'd be right! But the first mouse was invented in 1964. It took many years before the mouse would be widespread and commercialized. This is where agents are right now. OpenClaw gives you a tantalizing picture of where we're going. But if you've actually used OpenClaw, you know it's buggy, complicated, and is not smart enough to have it manage your money. It routinely ends up going bankrupt doing stupid shit. That will change. Right now OpenClaw is using models out of the box on tasks effectively outside of the training distribution. That's why it feels so broken. But none of the labs have RL'd against OpenClaw traces yet--and those traces are rich with signal. Once you see the first lab-released Claw-style model and harness, expect a huge jump in performance and consistency. Every lab has tons of OpenClaw data now, and they're all working on this, because they see how big the prize is. Expect this to take some time. Right now it's just the tinkerers playing around with this stuff. x402 is only doing a ~million in volume a day (and MPP much less), which shows that right now it's just tinkerers. But a next generation of models will likely be released within months. That will spell the end of the tinkering era and the start of early adopter phase. But even that will be early. The early majority will take years. This is a long-term story--the early adopters, the early majority, the late majority, and then the late adopters. But agents are coming for everyone, and they will change how all money moves. As the @cdixon mantra goes: what smart people are doing on their weekends and evenings, everyone will be doing in 10 years.

@ekrahm They are just telling the truth, but people don't believe them

Delusional confidence is such a superpower. Looking back, the founders I've worked with who went the furthest all started with socially unacceptable levels of self-belief

@kevinrose We also need to solve the payments and verifiable settlement problems for agents. The current methods (credit cards, paypal, Stripe, blockchains) don't work at the scale needed for agents.

Five years out, when billions of coding agents exist, software development will be largely solved. The traditional moat of “we have better engineers” disappears. Products will be copied, improved, and open-sourced almost instantly. Defensibility will shift away from code itself, toward distribution, data, brand, and community.

@voidsnam In programming languages, good syntax and type system design leads to seamless semantics. In the AI agentic world, this translates to good API design leads to seamless intelligent usage.

@RosuGrigore syntax design always takes the most time

Wadler's (simplified) law of language design: "Syntax takes twice longer than semantics to design" Funny how Wadler's law applies to agents: "API syntax takes twice longer than its semantics (understanding what you can do with it)"

@0xNairolf Crypto and AI are here to stay, but neither of them needs blockchains.

@0xNairolf Yep. Nobody needs blockchains. Neither crypto nor AI.

crypto needs ai ai needs crypto