> I don’t use windows, so most of this doesn’t affect me, but I do use GitHub and VSCode. Can anyone clarify, once and for all, whether “GitHub Copilot” and “VSCode Copilot” (sic?) are the same product? The documentation isn’t even clear, and it’s important because it affects billing. How do these two products interact and where do they NOT overlap?
There is no VSCode Copilot. There is Github Copilot integration inside VS Code.
Not parent but I wrote an article about how PDSs (the thing that hosts your data) control your signing and rotation keys. It's technically possible to self-host but as always, the default becomes the norm. 99.999% of users on ATProto host their data on BlueSky servers.
Permissioned data is probably the most fundamental, the part I looked most deeply into myself. People want privacy over blasting everything out to the internet for anyone to scrape. The public by default forced upon users is a bad choice. The purported benefits never materialized. Many of the atmo developers have this belief they can skip the network effects, grift the data and social graph for their own use.
Here's the User Intent proposal that is super easy to implement, yet they have been sitting on it since: https://github.com/bluesky-social/atproto/discussions/3617
This would have been at least a middle ground to permissioned data, as would have been personal private data (bsky prefs generalized).
After that money, which I see as less of a protocol thing. A protocol or platform has to enable the people to make way more money than itself, at least 10x. (1) Bluesky should have created subscriptions for their service, they wouldn't have needed the private equity had they. (2) Bluesky did more to block others making money than enable it. Graze was in talks with them to enable the creators using their feed system to make money, until Bluesky walked away. (3) Permissioned data would unlock monetization without blockchain.
Permissioned data is being worked on, but the commentary from Bluesky is not promising. (1) Nobody in ATProto has built a permission system (that I'm aware of) (2) Bluesky are proposing a very simplistic system. This will put burden on app developers and create opposition the credible exit philosophy.
Record history / editing. The former should be at the protocol level, the later on feature that is highly desired, possible today, but they resist with fervor.
Bluesky could have put way more funding into the ecosystem, especially in hindsight with the $100M they picked up just after peak. Now they are struggling and stepping on that ecosystem, re: replacing Graze instead of supporting and integrating them with their latest "ai" stunt.
Compare this to Hytale and what they are doing. Night and day.
The Bluesky team has also made several PR mistakes, upsetting their base, they are really tone deaf. Hope the waffles are tasty!
The PLC comes up a lot, and I understand the criticism, but it is also good enough for now and on the right trajectory, though the pace could be better, hut like much of the Bluesky development it has molasses in winter vibes. Long-term, multiple identity authorities can exist. Something like Handshake would have been ideal, another great project doomed by poor leadership.
Supporting delete is a good decision in my opinion, and likely a legal requirement. I also like that ATProto stuck a balance between decentralized and user experience. Properly federated systems are unlikely to appeal to the masses, re: blockchain.
Two other well designed parts of ATProto are how the algos and moderation work. Modular, composable, and anyone can participate. This would change with a properly permissioned protocol (zanzibar + macaroons imo) and encourage smaller social instead of big social.
Yes, because you know what all of the 200,000+ employees are doing in every wing and branch of the entire company.
Then again, Microsoft themselves directly dispute your statement:
Across the landscape of more than 750,000 devices in use at Microsoft, we support Windows, Android, iOS, and macOS devices. Windows devices account for approximately 60 percent of the total employee-device population, while iOS, Android, and macOS account for the rest. Of these devices, approximately 45 percent are personally owned employee devices, including phones and tablets. Our employees are empowered to access Microsoft data and tools using managed devices that enable them to be their most productive.
Not to mention that most app designers use OSX for the design tools, which means that there is going to be by default some bleed between the two systems on design choices alone.
> while iOS, Android, and macOS account for the rest. Of these devices, approximately 45 percent are
Pretty much everyone has an android or iOS device in their pocket. A lot of those devices are enrolled into Microsoft MDM in order to access email/teams/etc. These phones are part of the stats. Dev work in general is done on Windows boxes, unless you are in specific teams that have other requirements. Default is Windows, specifically Windows laptop.
Worst case somewhere around 50,000-150,000 tablets.
That leaves ~200,000 unaccounted for devices with only macOS on the table. I think the saturation is higher than you have experienced, although I'll give that it's entirely possible that the areas you worked in were not one of them.
Can you clarify - are you implying that BlueSky team made protocol hard on purpose, in order to "tell regular people that they are not allowed to participate"?
No, OP is saying that they have over-engineered the protocol, and that this acts as an *effective* barrier to participation, regardless of whether it was intended or not. Bluesky's protocol is focused on twitter-scale use-cases, where every node in the network needs to be able to see and process every other event from every other user in able to work properly. This fundamentally limits the people who can run a server to only the people who are able to operate at the same scale.
Email, RSS, blogs, even Mastodon protocol (it's not ActivityPub) scales better. Anything that only sends data between interested parties, instead of to everyone.
What can be stated without evidence can be dismissed without evidence. It is IMO pretty clear to me there is no substance to this post, without knowing anything about the author.
In general most such claims today are without substance, as they are made without any real metrics, and the metrics we actually need we just don't have. I.e. we need to quantify the technical debt of LLM code, how often it has errors relative to human-written code, and how critical / costly those errors are in each case relative to the cost of developer wages, and also need to be clear if the LLM usage is just boilerplate / webshit vs. on legacy codebases involving non-trivial logic and/or context, and whether e.g. the velocity / usefulness of the LLM-generated code decreases as the codebase grows, and etc.
Otherwise, anyone can make vague claims that might even be in earnest, only to have e.g. studies show that in fact the productivity is reduced, despite the developer "feeling" faster. Vague claims are useless at this point without concrete measurements and numbers.
> We’ll unpack why identical tools deliver ~0% lift in some orgs and 25%+ in others.
At https://youtu.be/JvosMkuNxF8?t=145 he says the median is 10% more productivity, and looking at the chart we can see a 19% increase for the top teams (from July 2025).
The paper this is based on doesn't seem to be available which is frustrating though!
I think you are quoting productivity measured before checking the code actually works and correcting it. After re-work productivity drops to 1%. Tinestamp 14:04.
In any case, IMHO I think AI SWE has happened in 3 phases:
Pre-Sonnet 3.7 (Feb 2025): Autocomplete worked.
Sonnet 3.7 to Codex 5.2/Opus 4.5 (Feb 2025-Nov 2025): Agentic coding started working, depending on your problem space, ambition and the model you chose
Post Opus 4.5 (Nov 2025): Agentic coding works in most circumstances
This study was published July 2025. For most of the study timeframe it isn't surprising to me that it was more trouble than it was worth.
But it's different now, so I'm not sure the conclusions are particularly relevant anymore.
As DHH pointed out: AI models are now good enough.
reply