Line-by-line streaming: who approved this and why?
The latest version introduced line-by-line text streaming. I assume whoever designed this was inspired by watching a dot-matrix printer from 1987, because the effect is identical: text crawls onto the screen line by line, the layout constantly shifts, and reading anything mid-response is simply not possible.
I genuinely cannot fathom what user problem this was meant to solve. The previous behavior — rendering text as it arrived, without turning the output area into a slot machine — worked perfectly well. This is a step backwards in usability presented as a feature.
Please provide an option to disable it. Or, alternatively, explain the reasoning — I'm open to being wrong, but right now this looks like a change made by someone who has never had to actually read their own tool's output.
12 Comments
Found 3 possible duplicate issues:
This issue will be automatically closed as a duplicate in 3 days.
🤖 Generated with Claude Code
It's pretty awful. Looking to downgrade but not even possible since using claude install native installer...
@bcherny please look at this
+1, very awful. can't scroll and read, streaming resets scroll
2.1.78 patch notes has it noted as a feature, did they add a config toggle to turn it on/off?
If not, we NEED this as a toggle if they're planning to keep this as a _"features"_
This is as bad as the flashing/jittering terminal behavior, I don't know who's idea this was to implement this.
Don't fix whats not broken guys...
actually it was introduced somewhere after 2.1.6x version. before claude responded with one big chunk/answer and it worked fine and I was happy with it. and it was easy to scroll and read. now it stream line by line, too fast, I am scrolling up to read it, scroll resets to top, I scroll again, again reset. so streaming output is totally unusable until it is finished
The response streaming change in v2.1.78 to line-by-line is a significant UX regression. Why even do this? Who does it help?
p.s. downgrading to v2.1.77 didn't help
Honestly, they need to stop with all these new features and changes and take some time ironing out the core UX.
They have data on users who use CC for 100s of hours, ask them for feedback directly.
I will happily take the survey if asked, please guys, refine the core experience.
https://code.claude.com/docs/en/changelog#2-1-78
Really unhappy with this change, it makes it much harder to read claude's responses with everything constantly shifting around on the screen.
You have GOT to be kidding me. Apparently there's a config fix for this but the config editor is broken! It's currently impossible to save any edits to the config 🤡 (bug report)
Reduce Motion config makes things better, 100%, thanks for pointing that out, but still not the same experience we had though
Your #35579 line stuck with me: "a change made by someone who has never had to actually read their own tool's output." The real complaint underneath "who approved this and why" is that something landed and no one owned the judgment.
So I'm curious: when an AI-generated change shows up in your tree, how do you decide it's safe to merge — read every line yourself, or is there a faster signal you trust? And is that line-by-line read something you actually want to do, or something you'd rather not have to?
No deck, no demo. I'm a solo founder building one thing: AI now ships ~30 changes overnight, and someone has to judge all 30 before merge — 10 min each and the day's gone. I build that reviewer a pre-made health report from last night so they decide merge-safe in minutes, no deep-dive. Does that sound useful, or like it'd get in your way?
— Shiyu (solo founder)
+1, a setting would be great to enable/disable this