Crypto: Solana, more than one operation in two serves no purpose. It's time to clean house.

Since yesterday, the new features in the Agave 4.2 update have been activating one by one on Solana, one of the largest cryptocurrency networks on the planet, ranking just behind Bitcoin and Ethereum. Nothing spectacular on screen. But there is one figure in there that I reread twice to make sure I hadn't looked at the wrong line.

During the week that ended on July 31, 51% of the operations carried out on this network were machines telling each other that they agreed. Not people sending money to one another. Just computers voting.

On the left, what people do. On the right, what machines write. Same notebook, same fees.

One operation in two is a machine raising its hand

Two minutes of explanation, and after that everything else falls into place.

A network like Solana resembles a large shared notebook. It records who sent what to whom. This notebook belongs to no one: it is copied identically onto thousands of computers spread around the world, called validators. Their job is simple to state and tedious to perform: constantly agree on the exact order of the lines. Because if two machines write different opinions, the transaction is considered false, malformed, or invalid. This is what protects the Solana network from fraudulent transactions.

The problem comes from the way these agreements are recorded today. Every vote from every validator is added to the notebook like an ordinary operation. It takes up space and incurs fees. So the notebook ends up half-filled with “ok,” “ok,” “agreed,” “ok.”

On the left, people. On the right, machines congratulating one another. The network pays for both.

The figures, measured during the last week of July: 1,620 votes per second, compared with 1,474 operations from real users. Machines talk more than customers. Imagine a supermarket checkout where one employee out of two does nothing but loudly confirm that the previous employee scanned the item correctly. It works, nobody makes a mistake, but you'll never get more people into the store.

What really changed yesterday

The version that has been activating since August 17 does not solve this problem yet. It does three other things, and they are very concrete.

The network's interval drops from 400 to 200 milliseconds. This duration corresponds to the window during which a machine can write in the notebook before handing over. It will be cut in half, but gradually: four 50-millisecond stages, with a safeguard that stops the reduction if too many windows begin to be missed. That's engineering prudence, and it's good to see.

Storage costs are falling by 90%. Keeping data in the notebook costs money because it takes up space for everyone, forever. The value used to set this rate drops from 6,960 to 696. In real money, the small account used to hold a token goes from about 16 cents to about 1.6 cents. Ten times cheaper to open the door.

And operations can be three times larger: 4,096 bytes instead of 1,232. A byte is roughly one character. So we move from an SMS to a page of text. Older formats continue to work; nothing breaks.

The biggest change is in a box

Here's the real subject. This version carries the code for Alpenglow, the biggest change ever attempted on this network, and that code is delivered switched off.

Alpenglow addresses the 51% problem. The idea can be summed up in one sentence: instead of recording their votes in the notebook, validators vote directly among themselves, by message, and a single compact signature summarizes thousands of votes at once. The notebook no longer sees any of it. About 75% of the space is freed up, and that space goes back to the people, to their transactions.

The other effect is even more impressive. Today, a transaction takes about 12.8 seconds to become irreversible, meaning impossible to cancel. That delay would be reduced to about 150 milliseconds. Eighty-five times faster. The system was proposed by Anza's team together with researchers from ETH Zurich, it was approved by validators by 98.27% in September 2025, and it has been running on a test network since May 11.

Twelve point eight seconds—that's how long it takes to read this caption. Twice.

A small detail that says a lot about how nervous the developers are: since August 5, a bounty of up to 50,000 SOL has been offered to anyone who finds a flaw in the four parts of the code concerned. The window closes tomorrow, August 19. When you put that kind of money on the table to have your own work broken, it means you are perfectly aware of what you are touching.

It's a bit like fixing an airplane engine in mid-flight!

Concretely, what does this change for you?

If you've never bought cryptocurrency in your life and have no desire to, you can turn the page here, but to me this is the future. If you're an investor and the Solana blockchain is giving you doe eyes, I think this change will polish its image even further.

What changes: the patience threshold. Twelve point eight seconds is a very long time when you're standing at a checkout with thirty people behind you. That delay blocks almost every everyday use. No one will pay for their bread with a system that forces them to stare at their shoes for thirteen seconds. At 150 milliseconds, the wait becomes invisible: it's the time it takes to blink, exactly the feeling of a contactless payment.

The terminal that dialed a phone number. Your grandmother remembers the sound very well.

The precedent exists, by the way, and it is telling. In the 1990s, paying by card at the baker's meant a terminal dialing a phone number, a little modem melody, and thirty seconds of silence at the counter. As a result, you only took out your card for large sums, and the baker would roll his eyes. The card itself hasn't changed since. It's the waiting that disappeared, and that's the only thing that means you can now pay for your baguette by placing your phone on a terminal.

Second concrete point: cost. When storage prices are divided by ten, the fees for small transactions go down. Sending three euros while paying one euro in fees is a joke, and that's what killed most everyday uses on these blockchains.

And what it doesn't change, I prefer to write in black and white: it doesn't make these currencies more stable, it says nothing about what they are worth, and it doesn't turn them into a payment method accepted by your butcher. It removes a technical obstacle and adds some serious advantages.

The promise is serious, and so are the reservations

The 150 milliseconds remain a stated target. This delay was measured in a laboratory and on a test network, not on the real network under full load, on a day when everyone rushes to the same place. That's exactly the difference between a car on a test bench and the same car stuck in a traffic jam in Brussels on a Friday evening.

Then there's the date. The target is October, with the next version. A target isn't a date, and on this kind of project, targets move. If it slips to December or March, that won't mean the project is dead; it will mean someone did their job properly and chose to delay it.

Finally, the real risk. Changing the way thousands of machines reach agreement is the most delicate operation one can attempt on a network of this kind. It's the kind of surgery that, when it goes wrong, doesn't make the service sluggish: it stops it. Solana knows something about that—the network has already experienced several complete outages in the past. Hence the 50-millisecond increments, hence the automatic safeguard, hence the 50,000 SOL offered to anyone who can break it. The whole setup says the same thing: we know perfectly well what we're getting into.

What I think

There is no promise of returns, no fashionable new token, no press release announcing the end of the banking system. There are people who looked at their own network, realized that half the work consisted of congratulating itself, and decided to get rid of it. It’s plumbing. It’s actually the very definition of plumbing.

And to be honest, I mostly like that 51% figure because it’s embarrassing. For years, this network’s official narrative highlighted its speed and number of operations per second, brandishing them like a record. Except that one operation out of every two in that record was the counter measuring itself. It takes a certain amount of courage to publish the figure that shows you’ve been telling yourself a bit of a story.

Now, we’ll see in October. Or December. Or next year, for that matter. But there’s no denying it: if Solana continues improving its blockchain like this, it’s going to start becoming indispensable.