2026 Atmosphere Report

This is the first ever Atmosphere report based on the “Where We’re At” survey. As such, there is no data from any prior year we can reference, and it’s not possible to take a look at trends and year-over-year changes. This report will thus focus on interesting infos from this years survey, as well as community sentiment expressed through the comment feature.

After a month of answering questions, the 2026 edition of “Where We’re At” has concluded! I’d like to thank everyone for participating. We ended up with 252 respondents from over 20 countries, with a very good distribution of experience. About 1/5th of the answers came from people new to the ecosystem, having building on ATproto for less than six months. A further 7% were veterans with more than three years of experience. And of course, there’s lots in-between these two groups!

This is the full report, taking a look at some of the data in detail and pointing out some interesting bits and pieces you might not spot immediately, or bringing up topics which were only commented on in free-form comments. For the data itself, and related comments, please check out the preamble and the data on the other pages that come after it. Basically: if you want the full story, please take a look at the data yourself!

What the data tells us

Let’s take a look at the more decisive parts of the survey.

Identity and hosting

Starting with the “Identity and hosting” section, the spread of Atmosphere account hosts as well as the number of users using their own domain lets us know the audience who answered this survey. We’re mostly looking at developers here, or at least tech-savvy people, which should be no surprise, considering this is a developer survey. What is surprising is the big chunk of Tranquil PDS users. Roughly 33% of self-hosters have chosen to not go with the reference implementation!

While the amount of people using did:plc as their DID of choice might look extreme at first, we have to remember that it is the main method, with did:web being newer and aimed at single users rather than bigger PDS implementations, so the spread here is no surprise.

Another part of the hosting category that is surprising is the amount of people not knowing about rotation keys. Since we are talking about a developer audience, educating them and non-developers on the ins and outs of account ownership might be worth it in the long run.

SDKs and languages

When it comes to SDKs and languages, TypeScript dominates the field. 70% of builders are using it! Its older sibling, JavaScript, comes in at a close second, just above Rust at third. As the label for that question points out, considering the official SDKs are written in TypeScript, this is not much of a surprise either. Though a lot more people prefer the atcute libraries, as the next block shows.

Over 50% of respondents have expressed either an interest or have used them before with a positive experience! Over in Rust land, jacquard takes the cake. One other thing that is notable is the amount of languages used to make ATproto apps. 30 Languages for such a small survey? Pretty impressive! In the same vein, a few respondents said that they’d like to see some libraries for their programming language of choice.

Though, no matter which language or library, the most frustrating part seems to remain OAuth and associated struggles, like outdated or missing documentation. Lexicon code generation comes in at a close second. Another noteworthy thing is that good amount of respondents also takes issue with the use of AI within the libraries they use, with comments being made multiple times that “AI slop” seems to take a precedent in them.

Infrastructure and tooling

Over in the “Infrastructure and tooling” section, we start off with Constellation being the most used service for reading data from the network, while also gathering the most eyes from people who haven’t used it to date. I’d call that a big win for community-run infrastructure! A common theme was users building static sites, and not wanting to write their own backend.

Another noteworthy thing is that nearly 50% of respondents self-host a PDS. Comparing that with the earlier 40% of respondents saying that their own account is hosted by Bluesky paints a very interesting picture: most people seemingly self-host or use Bluesky’s services, and only about 10% use different account hosts like Eurosky or Blacksky.

Moving down to the next card, we’ve got another win, this time for the homelab enthusiasts. Almost 60% of respondents host their apps on their own hardware, which is more than double of what Cloudflare was able to achieve! Talk about decentralization.

Last but not least, we’ve got third-party labelers taking the top spot for moderation tools, which might be due to them fulfilling different use-cases from the more general labeling alternatives.

Building on the protocol

To nobody’s surprise, the main issue for most people building on the protocol is OAuth, either due to the complexity of OAuth in general, lacking documentation, or the complexities of DPoP. Another big issue is local development friction. All of this combined leaves a lot of things to be desired on the developer experience side of things, which could be interpreted as a sign of a rapidly growing standard.

Speaking of rapidly growing standards, ATproto spaces are clearly the next big thing for builders, with over 85% of respondents expressing some amount of interest in them. Private data is in hot demand. As are apps for a new internet! Over 50% of respondents noted that their apps are meant to stay free, outright rejecting the idea of their project earning them money.

With only ten responses from users who have apps that earn them money, it’s not clear what the actual spread in terms of monetary success looks like for now, something that will surely become clearer in the coming months and years. Until then, the private funds of builders are likely what will keep the Atmosphere going, as is customary for now, with over 90% finding their own work on their ATproto projects.

Apps you use

While the apps section has been cut due to some naivity on my end (feel free to read the preamble for a quick overview), I’d still like to share the question that asked people which apps they’d like to see being built on ATproto, since it offers a very interesting insight: people want a lot of things to be built on the AT Protocol!

Reddit takes the top spot at 8.9%, almost double the percentage points of a marketplace. Starting from there, this question covers basically everything, from B2B apps and Google alternatives to music and video streaming. If you are looking for something to build, this list might be able to give you some ideas of what people want!

Open questions

Meanwhile, in the “Open Questions” section of the survey, the data is fairly self-explanatory. Thus, I’d like to highlight the sentiment of the comments that have been left on specific questions.

The question regarding a potential shutdown of Bluesky PBC was the most contentious one. While almost all of the people commenting on the matter argued that the network would probably survive, they also noted that the network would shrink dramatically, mostly due to users on the Bluesky-operated (Mushroom) PDSs suddenly losing their accounts. Some argued that the ownership of the Mushroom PDSs should be split up between different organizations, or to have a migration plan ready for these users in the event of a shutdown.

Having said that, about half of the commenting respondents remain optimistic that the Atmosphere is at a point where it could recover, although slowly, and with a much smaller userbase for the foreseeable future.

A fair amount of these comments also pointed out the PR behavior of Bluesky, which they noted comes across as controversial at best. Most were concerned with the actual marketing of the Atmosphere if the apps continued tying themselves to Bluesky as the biggest app, wondering how this will impact the health of the protocol long-term.

Moving on to infrastructure hosting costs, about half of the comments expressed optimism regarding cost efficiency and mostly believed that the infrastructure could be optimized to be less resource-intensive. The other half remarked that running anything close to the scale of a relay or the Bluesky AppView is nowhere near feasible, both in terms of hardware and cost requirements.

Another topic that received a lot of comments was the question about Bluesky PBC educating users about the network. As the stats themselves will tell you, only 5.9% of users thought that Bluesky was doing enough to educate users! Most comments pointed out accounts, how to migrate, OAuth (with bsky38 being the leading argument), and general knowledge about the Atmosphere.

Another few comments also noted that developers should be educated on the capabilities of ATproto, in the sense that a lot of possible apps are not being explored yet, since people are used to thinking in centralized platforms. Having said this, other users also commented that this should be the job of an ATproto foundation, or an equivalent organization, while Bluesky should focus on just microblogging.

Last but not least for the open questions, builders remarked that the hardest thing as of right now is user acquisition, with over 25% of respondents picking that answer. This could probably be associated with the missing user education, which came in third, right below funding and above private data.

To 2027 and beyond

That was a lot of data! I’m sure there’s many different conclusions to be drawn here, but for me, the main aspects were:

• The excitement about spaces,
• The concerns regarding Bluesky PBC,
• The issues with user aquisition & education,
• Funding,
• and the plethora of apps still waiting to be built.

Considering the amount of comments regarding Bluesky and their current relation to the network and community, I think that both sides should work to improve their standing, both in each others eyes and when looking at a more general audience.

A good amount of work is being done already, for example splitting the PLC into its own, seperate organization, which seems like an awesome starting point for having the network be more decentralized in the future. The IETF taking on ATproto as a standard is also a great sign for something huge in the future.

Looking into 2027, it’s hard to say how these topics will evolve, since we don’t have data from previous years to go off. But I remain optimistic that we will figure these things out together. We have a lot of amazing folks with skillsets that cover many different areas, so let’s put them to use!

Thank you for participating

That’s it for this year! I’d, again, like to thank everyone who participated. I’m aiming to re-do this survey next year and have already spoken to various people about this to ensure it will meet expectations. I’ve also been looking at a non-builder focused survey! Though both of those will happen next year, after Atmosphere Conf in April/May. Currently, I’m aiming for July/August at the earliest.

I hope the first edition was enjoyable, and the data was helpful. As I said at the start, I encourage you to take a look at the raw data if you’re interested, either by clicking through the site or by downloading the CSV file and taking a look at the codebook to get that parsed.

That’s it for now! If you have further questions, or simply want to keep up with new developments, please follow the Where We’re At account on Bluesky. Until next year!

- Lou