Aaron Ross Powell wrote a piece reacting to the recent TechCrunch article about Bluesky’s shrinking active user base, and it got me thinking too. I’d already been seeing the usual funeral procession online, Bluesky is shrinking, therefore Bluesky is failing, and that is a pretty narrow way to look at what’s actually happening.
TechCrunch isn’t wrong about the numbers, but the piece leads pretty heavily with the decline story. It does eventually get into the broader AT ecosystem, but by then the headline has already done what headlines do and as I mentioned people are off declaring Bluesky dead again.
Bluesky's active user base is down from its peak, and if you’re measuring Bluesky as a social media app, you should measure that. You should care about retention, daily use, monthly active users and whether people are sticking around.
What I’m not convinced of is that those numbers tell us very much about the health of AT Protocol. And the they don't, really.
Observers are still using the same measurements they've used for social media for twenty years. Monthly active users, daily active users, engagement, growth. Bigger numbers good, smaller numbers bad. That makes sense when the app and the network are basically the same thing.
Facebook owns the application, identity, social graph and data. Twitter (still won't call it X) does the same. Threads does the same.
But those of us building on the AT Protocol know that this isn't the case. We see a different picture, at least I do. We need to help others see that as well.
Bluesky is obviously the biggest application on AT and still has most of the load and visibility, but there is now a growing amount of activity underneath and around it that isn’t really captured by those looking at Bluesky's MAU graph.
There are independent PDSes running. People are hosting accounts outside Bluesky infrastructure. New AppViews are being built. Developers are defining their own Lexicons and storing things in repositories that have nothing to do with microblogging. Sovereign PDSes are online: @eurosky.social , @gandersocial.ca and others.
If I wanted to know whether AT Protocol was growing, I’d be looking at those numbers too.
How many sovereign or independently operated PDSes are standing up? How many accounts are actually moving to them? Does that offset the drop? How many applications are building their own AppViews? How many new Lexicons are being created, and how many are being used by more than one application?
Those are the numbers I’d rather be watching if we’re trying to figure out whether the protocol itself is growing.
The current numbers are messy, which is another reason I think we need better KPIs. Community analysis has found thousands of independently operated PDSes, while AT analytics sites are seeing thousands of third-party collections outside the standard Bluesky namespaces.
Some of those will be experiments. Some will disappear—I have a few of those already. Some probably haven’t been touched since somebody got excited one Saturday afternoon and then went back to whatever they were supposed to be doing (guilty, me again.)
That doesn’t bother me much. Early ecosystems are messy.
Raw account numbers bother me more because they’re easy to misuse. AT has a lot of spam and junk registrations, so waving around huge DID numbers doesn’t necessarily tell us anything useful either.
I’d rather look at what people are actually building. Look at the activity out side of the big app.
@standard.site is a good example.
@offprint.app , @leaflet.pub and @pckt.blog could all have created their own incompatible way of representing long-form publishing. Instead, they worked around a shared Lexicon. Other applications can now understand those records, and @bsky.app itself has added support for displaying them.
Nobody had to buy anybody. Nobody needed some special API partnership. They agreed on what the data should look like and built around it. That’s the kind of thing I’d count.
I’m seeing the same sort of discussion starting around books, which is particularly interesting to me.
Byline.pub is already in that space, using AT identity and structured author and book records instead of treating everything as another social post. @bookhive.buzz is coming at books from the reader side, with more apparently on the way. Cavu Reader is another application working around reading and discovery.
Now there is work happening around a shared books Lexicon.
I don’t want Byline to become Goodreads. I don’t particularly want anyone to build another Goodreads. That just gives us another giant silo.
On the AT Protocol, under the hood, Byline can concentrate on author identity and book information. BookHive can do social reading. Cavu can build the reader experience it wants. Another application can do book clubs. Another can build recommendations. Somebody else can decide they want to build the best science-fiction discovery engine on the network.
If the underlying Lexicons are shared, those applications don’t all need separate copies of the same data locked inside their own databases. An author shouldn’t have to recreate a book ten times. Just like a reader shouldn’t need a completely new identity every time they try another application. And an application shouldn’t be its own little island.
Anyhow, I digressed a little there, if Bluesky loses another million monthly active users next year, that is a problem for Bluesky.
But if, during the same year, hundreds more PDSes come online, new AppViews launch, shared Lexicons gain adoption and more active users start using the same AT identity across more independent services, I’m not sure you can reasonably describe AT Protocol as shrinking.
Again, you can say Bluesky is shrinking, but those are different measurements.
If I were building a KPI dashboard for AT Protocol, Bluesky active users would be on it. I wouldn’t ignore them. I’d also want independent PDS growth, migrations, accounts hosted outside Bluesky infrastructure, independent AppViews, relays and mirrors, applications using AT identity, records outside the app.bsky namespace, and Lexicons being used by multiple unrelated applications.
I’d also want some measure of how dependent the whole thing still is on Bluesky. That discussion has resonated on Gander and its Bluesky integration. How much integration do users want? How much of the Atmosphere can operate without routing everything back through Bluesky infrastructure?
Obviously Bluesky deserves a lot of credit for getting AT Protocol this far, and right now the ecosystem still depends heavily on it. We can't lose sight of that fact. But if the long-term goal is a genuinely distributed network, success should eventually mean Bluesky becomes less essential to every single part of it.
We’re already seeing pieces of that. Independent PDS hosting exists. Other AppViews exist. Shared Lexicons exist. Hubble is being developed as an independent public mirror—yes, with a grant from Bluesky.
None of that is especially exciting if all you’re looking for is a social-media growth chart. It's damn exciting if you are a developer or building on AT Protocol and it is interesting if you’re looking at infrastructure.
That’s where Aaron’s response to TechCrunch hit for me. I don’t think the answer is to dismiss Bluesky’s declining user numbers. I’m just not convinced they’re the best measure of whether the ecosystem around it is growing.
Meanwhile, there are numbers we barely talk about. Bluesky’s own CEO says the number of daily active apps on AT Protocol has doubled since the beginning of the year and is closing in on 1,000. SDK downloads have tripled. Blacksky has its own infrastructure and kept operating during a recent Bluesky outage. Eurosky is standing up European AT infrastructure and launched mu social. Gander Social is doing the same thing in Canada.
None of that necessarily adds another monthly active user to the Bluesky app.
I think some are still looking at AT Protocol through a Twitter-era measurement model. Or maybe the tech industry just hasn’t made the mental shift yet. We know how to measure platforms because we’ve spent twenty years doing it. We’re still figuring out how to measure a protocol deliberately designed not to become one giant platform.
Bluesky may flatten out while everything underneath it gets larger, more independent and more useful.
If that happens, staring at one MAU chart and declaring failure is going to miss an awful lot of what’s happening under the hood.
References:
We Needn't Have a One True Platform, @arronrosspowell.com
Bluesky’s active user base is shrinking as its focus expands beyond the app, TechCrunch
Toni Schneider interview, The Verge / Decoder, August 3, 2026
As a look back Beyond Bluesky: These are the apps building social experiences on the AT Protocol, TechCrunch