The latest Avalanche upgrade introduced Continuous Execution (CE), a new way to improve the performance of blockchains by taking advantage of parallelism. It’s exciting technical stuff that you should know about so you understand what makes Avalanche so fast, or if you're not using Avalanche, what you're missing out on. Avalanche has always had an architecture that takes advantage of coarse-grained parallelism. Namely, Avalanche L1s capture parallelism via separate chains, where unrelated groups of activities can execute independently of each other, without having to wait for or congest shared resources. This was a huge step forward when we introduced it 6 years ago. Compared to older, single-chain systems, this multichain architecture enabled Avalanche to offer infinite capacity. A game doesn't have to compete with DeFi on Avalanche, whereas on single chain systems, even an NFT offering would cause problems by creating congestion and bumping transaction fees. While this took advantage of the largest source of parallelism and enabled Avalanche to be one of the fastest chains in the first place, there are further opportunities for improvement, and Continuous Execution taps into the next greatest source of parallelism, namely, the overlap between consensus and execution within a single chain. The two most expensive activities in any blockchain are consensus, that is, the work required to get all of the participants to agree on what happened and when, and execution, the work required to actually process those transactions and update the state of the system. Traditionally, these two activities have taken turns. Continuous Execution allows Avalanche to overlap them. While the system is achieving consensus on one set of transactions, it is simultaneously executing previously agreed-upon transactions. Think of the chain as a jet. Execution is flight, transactions are the fuel, and consensus is refueling. Before CE, the jet had to land every time it needed to refuel. The chain would execute, stop to perform consensus work, execute again, then stop again. Continuous Execution changes that. Now the jet can refuel while it is still in the air. Consensus and execution happen concurrently, so the chain no longer has to constantly switch between the two. The flight becomes continuous, with no unnecessary interruptions or mode switching. To put this in perspective, before CE, Avalanche validators were spending somewhere between 20 to 60 milliseconds per block on verification, sometimes even as high as 150ms. CE dropped that down to an average of 2 milliseconds of wall clock time with negligible variance. And this is especially important because verification happens at each step during gossip dissemination. It isn’t simply 40 milliseconds saved once. That time was being incurred on every hop as a block spread through the network. There is also a deeper technical point here. The work that previously had to happen in consensus goes from potentially unbounded computation to linear in the number of transactions. We achieve this by taking advantage of the monotonicity of gas prices under maximum load, so the execution thread is never forced to pause due to estimation errors. We’ve always prided ourselves on bringing the best of science to blockchain technology, on being pragmatic, and on attacking the highest-yield opportunities first. With Continuous Execution built on top of the multi-threaded Avalanche architecture, Avalanche has taken advantage of the two biggest sources of parallelism: coarse-grained parallelism across separate chains, and the overlap between consensus and execution within a single chain. For Avalanche, consensus and execution no longer need to operate in bursts, constantly stopping and waiting on each other, it just keeps going with no dead time.

Thank you to everyone who came to Avalanche Summit this week. I felt a tremendous amount of gratitude while walking through the venue and seeing what everyone was building, talking with the community, and seeing how far we have come. To old and new friends from across the world: I hope everyone came away feeling as energized and inspired by what we are doing as I did. My sincerest appreciation to our marketing team for putting on such an amazing conference. See you next time!

The reason cross-border payments are slow isn’t the wire. It’s the pre-funding requirement. Today, banks have to park billions of dollars in dormant accounts around the world so payments can settle when needed. That capital sits idle instead of being put to work. When settlement is guaranteed in seconds at the protocol level, that requirement largely disappears. Capital can remain productive until the moment a payment actually needs to move. That’s what the Avalanche Payments Collective is building toward.

Ten years ago, I typed out a warning in real time in Google docs while hundreds of anonymous wombats and chupacabras watched over my shoulder. Three weeks later a hacker drained $60M, 15% of all ether was at risk, and Ethereum forked to survive. We called it before it happened. A decade of audits, formal verification, and hard lessons later, I think we've made enough progress to get it right the next time. I shared my story in @FortuneMagazine, out today.