Does anyone else hate reading AI summaries of code? Code can be pithy, but at least its terse compared to prose. When you add how verbose LLMs can be, I often end up reading a paragraph to explain a few lines. Or the opposite happens where the summary skips important edge cases or criteria. "You're right, X also does Y. I missed that in my initial analysis," is much too common of a phrase.
I like the idea of using LLMs to transform code into something more readable, and vice versa. I am not sure if meandering paragraphs and linear lists are the best targets.
I wonder if a productive avenue might be "doing aspect-oriented programming in reverse": asking the LLM not to "summarize" the code per se, but rather to "clarify" it by transforming it into what a programming blog post would call a "toy example" of what the code is doing, by stripping out all the (non-semantic) error-handling, logging, metrics incrementing, etc — all the things that you might treat as their own "aspects" under AOP.
The "language model" aspect shows IMO, at least for someone who grew up with the material of 90s and 2000s where we describe things more in mechanical, engineering terms. When gpt4 landed it was somehow amazing to see the output, but yeah nowadays I keep skimming through the explanation. It's like an intermediate dev who has nothing else to do but create long sentences to describe what could be simpler.
I don't like the default explanations. But I prompt for small code snippets with explanations of a problem and the solution. This is in the context of extending features, fixing bugs, reviewing new code etc. I still skim the code, but it's nice to have a somewhat thoughtful overview of the key points like database schema, API spec, algorithm or abstraction. It makes it easier to skim a large diff without feeling lost. It also quite often catches some weird choices that might slip through it not carefully reasoning about the code. For me it is hard to understand code I didn't write myself so I have landed on this workflow.
I use AI explainers a lot in my own projects and have skills that give outputs I like. I'll even pipe the output into NotebookLM to make a video sometimes. This is actually a decent way to grok complex subsystems (YMMV).
What I don't care for is the pretty-yet-verbose HTML outputs that I see from some teams. They can be way too long and it's often the case that the engineer who made it didn't read the whole thing.
I've found that I don't mind getting a lot of text back from an LLM, when I was the one who prompted it. I can easily enough let my eyes flit around in the text and figure out what I need to, and I expect that putting all that text in the context window will help with the rest of the conversation.
It's when that text gets copied and pasted into a blog, or a PR, etc. that it really galls.
I'm a bit confused by your use of 'pithy, but at least its terse'. Don't pithy and terse mostly mean similar things? Here it seems like you're contrasting them. Not trying to be pedantic, apologies if it comes off that way.
Your AI summary is raw output, like uncooked mince meat.
Ask for a rewriting pass! Get it to edit for word count, remove redundancies, remove run-on sentences, and change sentence structure to use active voice.
You wouldn't ship your first draft - why let AI do it?
Right, the value is not so clear of having multiple people wade through lengthy LLM responses. Part of the value that a human in the loop adds is figuring out what's correct and important in the response and passing on a more polished result.
We need a new specification language for summaries. I think just a (pseudo?-)programming language’s API (modules, interfaces, method signatures), but more concise and descriptive than Java, maybe Haskell or OCaml minus quirks (…the academics were this close to an ideal syntax, maybe they should’ve gave up those languages’ LR parsers…)
Have been following this for a while, but honestly still not exactly clear to me who/what this tool is meant for
After all, we already have safe conccurent edits, etc via regular git tooling
To me it this seems a bit backwards looking. Increasingly code review is more on verifying the functional requirements, and less about reading every line an agent has written (or watching it write those lines live)
This is intriguing. The two relevant features seem to be 1) realtime collaborative multiplayer conversations and 2) conversation-as-document - basically, letting you comment inline in an agent conversation.
For (1), the main value I'd see is in mentoring junior engineers or less technical contributors on a team. If someone puts up a PR with sloppy results, you could actually jump into the thread that produced that PR and see how the results came about, or even coach that contributor on how to do better next time. Also might make it easier to hand off work from one person to another - right now most coding agent sessions are user-local.
On (2), I frequently find myself consuming agents' gigantic text responses and tediously writing 8-bullet-point responses to guide them. It's pretty exhausting. I could see inline comments providing much better ergonomics.
All that being said, Zed has largely fallen out of the conversation for "agentic coding tools", and so this feels like their attempt at creating something like the Cursor Agents Window, Codex, or Claude Code. These two features seem compelling, and I understand they're even compatible with other coding harnesses. But I don't know if there's enough there to have a defensible product; if these features are excellent, others will clone them eventually.
> On (2), I do find myself consuming agents' gigantic text responses, writing 8-bullet-point responses to guide them, then iterating that way. It's pretty exhausting. I could see inline comments providing much better ergonomics.
For your 2 plannotator [1] has a nice way to annotate agent messages or documents and its integrated into the agent chat (codex app also adopted the pattern)
It's a solid experience so far (have been testing the alpha). Being able to work across model providers in parallel is my preference, rather than being stuck in Codex desktop, for example.
For me it's probably time to learn to use vim or something similar. I'm the only one that likes coding manually, without all this AI shit? I'm faster writing code myself than trying to explain to artificial stupidity what I intend to do and correct its output. And if I write it myself the code just works, if it's written by the AI it has a ton of bugs. I don't see any value in AI generated code that is not mere code for doing experiments that you don't care about.
Hey me too. This is my morning Zen moment refactoring and cleaning some code before I go back to the agent to earn money. One of the only things that helps me to relax is to do some manual coding...
It seems reasonable that a professional-tier of software developers will have tools distinct from everyone else. No matter what the percentage looks like, experts will still be a thing that needs different things.
I don't quite get it - it reminds me of using Slack as the decision making places, sure it works for hashing out the details, but it's not great for being the decision record store. I'm not sure what is the value of preserving hundreds of lines of conversations about how code came to be if the code keeps changing? What happens in 5 years? Do I have to read the whole transcript just to understand what's going on? AI will summarize it for me?
This feels like engaging with a prog lang community for the first time, and answer to a basic question is "this has been covered before, read the IRC chat history"
It's not like Pull Requests are perfect, and with discipline they do the job well - but this looks like a step in the wrong direction.
If one builds something with an agent, the process itself is largely opaque. Preserving the transcript shows the reasoning of the agent and the decision process of the builder. This way everyone's on the same page instead of re-prompting same CLAUDE.md but with different results, and/or going through similar work over and over again.
I understand the point, but shouldn't the code be documented and well structured so that the intent is clear without looking at the discussion around changes?
I'm working on a legacy system right now - I'd commit several felonies just to have well documented code, I don't really care about all the conversations that the original team had to decide what they eventually did.
This does look legitimately exciting. A surprisingly large pain point when doing agentic work has been commenting on something in a larger plan document. I always find myself summarizing the surrounding text to contextualize a comment when all I really want to do is highlight and click "add comment".
Unrelatedly, I have been looking at Zed to centralize my agentic work at $job, where we use different API keys per project to better attribute spend and control model availability based on per-project data protection controls. All of the standard UIs I've tried for this don't really work, but the CLIs mostly do. Using ACP in Zed I was able to bridge that into the UI world and I'm quite happy with it.
I signed up for the beta and I look forward to trying it.
Thanks for pointing that out. I tried the Codex App before but didn't see that.
I want to like the app, but I'm not sure if it actually fits well to my use case. With my current setup I need to fully quit and restart it if I want to work on a different project (to get it to use a different API key) and I cannot work in two threads belonging to different projects at the same time.
Also I generally prefer to handle sandbox the entire agent and disable its internal safeguards. Beside any doubts regarding the quality of the relevant sandbox implementations, I have just gotten my projects to work better this way.
Directly annotating parts of the convo is the feature that really unlocked Delta for me (have been testing the alpha). So much easier than trying to explain to the agent what I'm responding to in their giant text blob.
Honest question, do any of you actually use this type of multiplayer feature in editors? I have worked on small, medium, large and FAANG(Google and Amazon) companies. Never have I seen anyone, even talking about coding with somebody else in the same editor.
So, DeltaDB is not for Zed, but for an entirely new product? At least it won't bloat the editor...
Nevertheless another Zed post, another plea to focus on basics https://github.com/zed-industries/zed/discussions/54150 [how can developers work with agents in zed when developers cannot see files agents create] best luck to Delta but please do not neglect the text editor!
I think the point is to incorporate it in the end in to Zed. Becoming the new thread panel or something. But it's just not something that is easy to fiddle with inside a working editor that many of us rely on to make some money.
It's not a full IDE but you can see the files and docs mentioned and created in the Delta chats. I often will pop open the right tab in Delta to check that.
I feel like for a little bit I was worried about Zed's ability to stay up to date in the cli coding agent-focused market, but it seems like every time they publish a new product release, it's always very compelling. I feel like they're one of the more promising companies innovating in the space of human-AI interaction, and I'm very excited to see where they go from here. I hope they launch a web-accessible or mobile-accessible app one day. I've been enjoying being able to code from Termux on my phone over SSH or using any of the browser-native orchestration systems.
Delta.dev is the same Rust application as Delta, just compiled to WebAssembly and rendered through WebGL. So it's already web accessible, and can naturally extend to mobile platforms as well!
As an indie dev, I'm not really in a position to judge Delta's potential for large-scale collaboration. Maybe Delta will end up being great for teams of all sizes. But I really wish they'd put more care into Zed's core editing experience. In my opinion, Zed has so many bugs and rough edges that it's hard to consider it software that deserves a 1.0 version number. Also, there's still no timeline in sight for GPUI to become a standalone project...
Would be nice if Zed announced anything related to what Zed actually is: a code editor.
This thing is fairly new in software years and has already stagnated. Bugs, glitches and rough edges all around. Almost feels like abandonware.
I first heard of Zed back when it was “the code editor that respected your choices”, you could turn off AI entirely with a single switch and all that… guess it was too much of a pipe dream to hope it would continue down that road.
This thing is now just a white-label version of Visual Code. Nothing more. Nothing less. Same software, different case.
Just because they started out as a code editor doesn't mean it has to stay that way. I was looking forward to a more modern Sublime Text in Zed but I also know that nobody is going to make money off a simple code editor.
> This thing is now just a white-label version of Visual Code.
That's a wild hot take, and pretty ignorant to the people working on that for years.
I have been testing the alpha for the last few weeks. It's very good, especially compared to tools like Conductor. Very bullish on where Zed will take Delta!
Would be cool to have a single-user ADE tool like that from the Zed folks - because Kepler focuses a bit too much on the ticketing integration and Paseo sometimes has issues with sub-agents.
I can imagine this being useful, but unlike Zed, this looks like it'll come with some fairly significant lock-in. Great for the investors, but it makes me hesitant...
I see Zed abandoned their principles regarding AI. Shame, I really do like the editor. Any decent alternatives that aren't VSCode or a Jetbrains product?
You know it’s ironic that the first featured post showing up on the bottom is one where Zed proudly proclaims to not be doing AI just for the money that’s in it, and all posts that follow it are about AI too…
Does Zed even remember it is a code editor? Or are these guys just interested in building more AI slop to add to the never-ending pile? What even is Zed supposed to be at this point?
I hate to be that guy who complains about style over substance, but man, is that font they use for body text ugly. Not just the glyphs themselves, but the spacing is way off, too.
I like the idea of using LLMs to transform code into something more readable, and vice versa. I am not sure if meandering paragraphs and linear lists are the best targets.
What I don't care for is the pretty-yet-verbose HTML outputs that I see from some teams. They can be way too long and it's often the case that the engineer who made it didn't read the whole thing.
It's when that text gets copied and pasted into a blog, or a PR, etc. that it really galls.
Ask for a rewriting pass! Get it to edit for word count, remove redundancies, remove run-on sentences, and change sentence structure to use active voice.
You wouldn't ship your first draft - why let AI do it?
First attempt: mountains of meta conversation and not answering the original question.
Second attempt: shorter and more concise, cuts out some of the nonsense, but terribly written.
Third attempt: goes off the rails, misunderstands what you are asking and tries pushing the work onto you.
Fourth attempt: finally something that's concise, reasonably well written, to the point and passable.
After all, we already have safe conccurent edits, etc via regular git tooling
To me it this seems a bit backwards looking. Increasingly code review is more on verifying the functional requirements, and less about reading every line an agent has written (or watching it write those lines live)
For (1), the main value I'd see is in mentoring junior engineers or less technical contributors on a team. If someone puts up a PR with sloppy results, you could actually jump into the thread that produced that PR and see how the results came about, or even coach that contributor on how to do better next time. Also might make it easier to hand off work from one person to another - right now most coding agent sessions are user-local.
On (2), I frequently find myself consuming agents' gigantic text responses and tediously writing 8-bullet-point responses to guide them. It's pretty exhausting. I could see inline comments providing much better ergonomics.
All that being said, Zed has largely fallen out of the conversation for "agentic coding tools", and so this feels like their attempt at creating something like the Cursor Agents Window, Codex, or Claude Code. These two features seem compelling, and I understand they're even compatible with other coding harnesses. But I don't know if there's enough there to have a defensible product; if these features are excellent, others will clone them eventually.
Regardless, would love to give this a shot!
For your 2 plannotator [1] has a nice way to annotate agent messages or documents and its integrated into the agent chat (codex app also adopted the pattern)
[1] https://github.com/backnotprop/plannotator
But a lot has changed in those 12 months.
Frontier models and coding agents have advanced so much that I don't really see much value in this anymore.
Not sure the DeltaDB based features really add anything significant compared to the alternatives.
I reckon the game here has to be adding a service that stores the data and runs agent sessions?
Here they've moved the AI chat to the center and code editor has been pushed to the side.
This feels like engaging with a prog lang community for the first time, and answer to a basic question is "this has been covered before, read the IRC chat history"
It's not like Pull Requests are perfect, and with discipline they do the job well - but this looks like a step in the wrong direction.
I'm working on a legacy system right now - I'd commit several felonies just to have well documented code, I don't really care about all the conversations that the original team had to decide what they eventually did.
Maybe I'm just too old and don't get it.
Unrelatedly, I have been looking at Zed to centralize my agentic work at $job, where we use different API keys per project to better attribute spend and control model availability based on per-project data protection controls. All of the standard UIs I've tried for this don't really work, but the CLIs mostly do. Using ACP in Zed I was able to bridge that into the UI world and I'm quite happy with it.
I signed up for the beta and I look forward to trying it.
The interesting bit is this type of UI doesn't fit very well in any of the existing terminal TUIs, I have only seen it enabled working well in GUIs.
I want to like the app, but I'm not sure if it actually fits well to my use case. With my current setup I need to fully quit and restart it if I want to work on a different project (to get it to use a different API key) and I cannot work in two threads belonging to different projects at the same time.
Also I generally prefer to handle sandbox the entire agent and disable its internal safeguards. Beside any doubts regarding the quality of the relevant sandbox implementations, I have just gotten my projects to work better this way.
Edit: grammar
Nevertheless another Zed post, another plea to focus on basics https://github.com/zed-industries/zed/discussions/54150 [how can developers work with agents in zed when developers cannot see files agents create] best luck to Delta but please do not neglect the text editor!
https://podcasts.apple.com/us/podcast/syntax-tasty-web-devel...
(i worked on this)
This thing is fairly new in software years and has already stagnated. Bugs, glitches and rough edges all around. Almost feels like abandonware.
I first heard of Zed back when it was “the code editor that respected your choices”, you could turn off AI entirely with a single switch and all that… guess it was too much of a pipe dream to hope it would continue down that road.
This thing is now just a white-label version of Visual Code. Nothing more. Nothing less. Same software, different case.
What a shame.
> This thing is now just a white-label version of Visual Code.
That's a wild hot take, and pretty ignorant to the people working on that for years.
Last week, 300+ comments
https://news.ycombinator.com/item?id=49187256
Does Zed even remember it is a code editor? Or are these guys just interested in building more AI slop to add to the never-ending pile? What even is Zed supposed to be at this point?