> In certain markets, we use conversation data to train the generative AI models in Copilot, unless you choose to opt-out of such training.
"Build me a SaaS platform exactly like ____"
If agents become as good at long running tasks as we're told they will do by giving Microsoft access to your codebase and inner business processes to give to anyone that wants to the ability to clone your business.
That might end up being inevitable but I see no reason to accelerate that.
The most reasonable reading of that description is that VSCode itself is open source, not that it is only intended for editing open source software. Furthermore, nothing in the license suggests that. If that was their intent they were very much not clear about communicating it.
They still aren't honest about the fact that the official VSCode distribution isn't fully open source because, for example: you can't have VSCodium connect to the official plugin repository. It also isn't the only editor with AI integration, and more specifically these systems use LLMs.
It would therefore be more honest to say that VSCode is "a visible source LLM code editor".
That doesn’t make VSCode not open source. It’s fair that third party integrations like the extension store, or some extensions, are not open source. That’s what they are, extensions.
It does seem to me like Microsoft (and every other company developing AI models) is doing so at a loss, and a signifigant one.
Is the play here to get everyone hooked on AI and then jack up the price to make a profit?
If so, I worry about Junior devs in particular, who have never developed the skills to write software themselves, suddenly finding themselves being "cut off" from their AI dealer
Or people generally who outsource their thinking to AI, forget how to do things for themselves, and suddenly face a big bill!
This is 1000% the play, it's the only one that actually works out (for _some_ vendors). Extra fun when you've let go of all your actual experienced engineers and then the squeeze comes.
Well, I don't know what to think anymore. I use LLMs in my work, but I'm not comfortable signing my name on something I don't fully stand by (akin to the responsibility of a lawyer's signature).
It's December 27th, 2025 and I'm not supposed to be thinking about my future*. I'm supposed to spend time with my family and enjoy that. Yet here I sit mulching on this.
* I didn't add 'as a Software Engineer', because I wouldn't know as what else.
It’s a rote observation at this point, but: there’s a clear discrepancy between the demonstrated value of LLMs (which is, to be clear, significant!) and the aggressive manner in which Microsoft has introduced them into their products. The latter is what you do when you can’t demonstrate value, and it produces worse outcomes by design (because it starts from the assumption that users need to be given the stick instead of the carrot).
There have been a lot of recent changes to VS Code that feel like this: the Copilot pane has been refactored to take up more space and behave less like other composed panes in the window; the integrated terminal now does overly clever and brittle things to introduce suggestions in REPLs like Python’s. Those kinds of changes have pushed me more to Zed recently, which has all of the same AI features but without the user hostility.
I think a lot of this is just crappy bonus and incentive structures.
The execs want to they're using+selling AI, the investors want to believe AI can theoretically fire all the workers/drop your fixed costs, and the middle managers need to justify that they're on it by myopically pushing out features that increase the AI adoption metric.
The rushed push of AI features obviously trains your users that your AI is useless crap that just gets in the way. If you're going to do it's, make it limited and high quality first.
Yeah, completely agreed. It just seems like such a funny place to have those kinds of perverted incentives, given that this stuff is actually kind of useful and clearly has market fit!
I use AI a ton and I pay for Claude happily. They've found an incredibly valuable niche and built the best products for it. I almost fail to see what value an AI editor has in comparison.
I've released a number of AI fearures at work, but they're focused on being good at one specific thing.
I love vscodium but more and more I worry about how Microsoft is effecting it down stream. To the point where I'm actively looking into making my own editor. I'm putting it off for now but I'll probably start playing around with Theia and Codemirror on the side just in case.
I'd love to move back to it (or rather, use it for dev work beyond opening large log files to search for things), or atleast have it as a backup for vscode's inevitable enshittification.
same boat. I switched to codium mostly out of purity from AI, and I'd really like it to stay that way, while still getting other QoL improvements. I'm pretty concerned that there's not enough to justify the niche, though.
If I have to bet, I will absolutely go for MS enshitifying it beyond reasonable usability, in one way or another, more soon than later.
Making an editor is anice endeavor. But there are plenty of, which are extremely well developed, open source, in many directions, emacs and vim the most prominent. But many others out there.
Emacs and Vim are terminal based though. So nice things like scroll bars, tabs, drag and drop etc. might be available as hacks but will disappoint in the ways in which they fail to work like a actual GUI interface. I'm also not a fan of model text editors.
For open source GUI text editors there sadly aren't many that match the feature and polish of vscode.
While Emacs can run in a terminal, it is more widely used as a GUI application that can render images, PDFs, variable-pitch fonts, handle mouse support (drag-and-drop, menus, scrollbars), and even work on touchscreens such as on Android [1].
You are right that VS Code has a "nicer" out of the box UX (this is subjective of course), but Emacs offers a malleable environment. In VS Code, you are limited to what the APIs the developers decided to expose. If you want a specific behavior that isn't supported, you either fork the editor or create a feature request ticket and wait for someone to prioritize it. In Emacs, because you have full access to the internal runtime, you can implement that feature yourself in a couple of lines of Lisp.
Thanks, it's been so very long since I've tried emacs. I remember I didn't like how it looked. So I used vim instead. (There was no vscode back then.) So I never did give it much of a try.
Emacs might be a solid editor choice but my intuition is that it probably won't be worth it for the same reason LiteXL wasn't for me. If I do work on adding features to my editor I think I'd be more comfortable doing it in js, html and css. And if possible I'd rather start with a base that's mostly where I want it to be. Trying to turn emacs into vscode sounds like way more of a project than turning Theia or CodeMirror into vscode.
Actually there are plenty of packages already which can near Emacs to VSC or Sublime in look and feel, and imho go circles around the 2 in functionality.
I don’t know them, because I do not like VSC and co. I just have a friend, and when I see his Emacs looks like sublime. Is some work to get it to look like that.
My guess is they're vastly underestimating the time and effort that would take, however, I understand the motivation somewhat, as there's no guarantee that whatever alternative to VSCode you settle on won't also eventually go all-in on AI. For example, KDevelop is planning on heavy AI integration soon.
This type of branding makes it impossible to find products lately.
I was recently looking for embedded analytics platforms (and was willing to pay), but the search became incredibly frustrating as every database or analytics tool now brands itself as some AI first thing. The landing pages no longer help me figure out what they do, which I guess is good for raising investor money but I'm sure it can't be good for real sales.
I hope that soon the mania can end and we can get useful branding again.
VSCode to me is better branded as the editor with the best plugin ecosystem around. The AI features should just be plugins to an incredibly flexible editor. But I know MS wants to sell subscriptions like windsurf and cursor.
I guess there's a lot of pressure from Cursor and Google's Antigravity. Also with Zed you can bring your own API key which VS Code didn't support for a long time.
This pivot sounds like VS Code is moving from a text editor to a thin client for AI services that Microsoft wants to push. It is one more step towards a future where our development tools (just like everything else on our computers these days) are just thin clients/wrappers around SaaS.
Emacs remains the antidote to this. I use Emacs because I want to remain the architect of my development environment, not become the consumer of a telemetry-gathering platform architected by PMs at a big tech company. It is also an absolute joy to use an environment that provides you with the same amount of power as the core maintainers, allowing you to fully inspect and modify the system even while it is running.
My path in the emacs/vi divide forked a lifetime ago, and emacs is so fundamentally different that it was never worth sacrificing the massive productivity vim gives me to dip back into emacs
But maybe that should change. I like vscode for when I need more IDE features than I care to cobble together with plugins.
I don’t need another subscription in my life. Especially for anything I rely on.
I was more familiar with Vim bindings and relied on Vim emulation layers in various IDEs before I moved to Emacs. Evil mode and Doom made the jump possible without sacrificing too much productivity. With Evil, I didn't have to retrain my muscle memory and with Doom I didn't have to cobble together a functional config from scratch.
After a couple of months of using Doom, I felt comfortable enough to roll my own config which also helped me better understand how things worked at a lower level. More interestingly, after a couple of years, I transitioned from Evil to standard Emacs bindings as that felt better integrated with the rest of Emacs.
My path was similar, except with Spacemacs, which has excellent Vim-like modal key bindings. I've been using a custom config for many years now, but evil-mode has been a crucial part of my setup. A modal interface is simply easier to use and more intuitive than twisting your fingers to hit complex key chords. I use Vi mode in shells, TUI programs, REPLs, anywhere Readline is supported, etc.
So Emacs+Vim is the best of both worlds. You get the infinite extensibility of Emacs and a sane(-ish) programming language, with the superior editing and command interface. The beauty of Emacs is that it really doesn't matter how you use it. For some modes you may need to override a keymap, or use a package like evil-collection, but most behave well OOB IME.
I would not be surprised if the market share breakdown is similar to browsers (eg 70+ percent - more if you ignore that safari is the only real option on iOS).
VSCode has slowly been getting more and more bloated, but the alternatives are all very meh or are missing crucial extensions.
If you do embedded development, things like https://platformio.org/platformio-ide, but also smaller, nice to have extensions for auto-deploying code to cloud providers, etc.
To me that sounds like claiming Arduino IDE is a "crucial extension". Their website[1] lists a bunch of IDEs where it can be integrated, so I wouldn't call it missing. That said both of these are hobbyist toys to make it more approachable and embedded development was fine long before VSCode, they're in no way "crucial".
I loved vscode, but recently performance has been horrible like during right our in middle of longer agentic development. Call out to zed, please include Jupyter notebook views. Only reason i use vscode is for opening ipynb files.
Cursor has been annoying me lately with their updates breaking ever further away from vscode UI. Might give copilot another shot. Needs a plan mode though, it really is necessary for complex operations.
Jumped back to it to try seeing how functional it'd be as something more than than large logfile explorer.
Package control is still only in the command palette. If you want to explore what's on offer you have to do so on the actual package control site.
Managed to get LSP + intelephense installed so I have good PHP parsing (Other LSP providers appear to be available)... but stuck at the moment trying to get an intellisense analogue setup... Doesn't show up in package control in the program despite showing up on the site.
So right now I have syntax highlighting and errors flagged for a php file... but I don't have anything that can take the fact the class is missing several methods from the interfaces, and stub them out in a few keystrokes.
Im glad i moved from vscode to neovim last year, and since one month i’ve switched over to emacs, running doom (which gives you vim commands and much more).
I’ve set up LSPs, completions, etc and although one needs to read up a little bit at first, i feel that this could finally be a stable platform/ide for once, and i wouldnt need to jump ship every couple of years because of some enshittification.
AI-code editor or AI code-editor? Future versions may include traversing gigabytes of code, supervising hundreds of agents, and peer-to-peer (P2P) content-addressed caching.
Someday, could right-click a dependency and click "Zero dep," and it updates with a library integrated with the app. Stored in the cloud, other users benefit from the same generated output.
Apps become instances consuming them, the thinnest crust around various baked libs (mantle) or triggering changes in the molten core.
> In certain markets, we use conversation data to train the generative AI models in Copilot, unless you choose to opt-out of such training.
"Build me a SaaS platform exactly like ____"
If agents become as good at long running tasks as we're told they will do by giving Microsoft access to your codebase and inner business processes to give to anyone that wants to the ability to clone your business.
That might end up being inevitable but I see no reason to accelerate that.
reply