To me this (coupled with their DeltaDB announcement[0]) feels like Zed trying to get out of the business that Cursor is in, and maybe even kicking them on the way out, if the protocol helps bring-your-own-agent take off. These agent tools are already not particularly sticky, especially when they are outside the editor. I don't use Cursor or Zed full time, but I have heard that the most compelling bit is the tab completion, which isn't even part of the agent stuff. I'm sure it's nice, but it's very hard to imagine that being a moat for an editor.
I think I agree with the implicit judgment from the Zed team that it's too early to try to lock people in and capture value, and the best way to build long-term user loyalty is to just be the best editor and let people work however they want to work. On top of that, it is not a great use of their dev time to iterate on the details of an agentic coding tool when there are 10 other ones doing the same thing, and any prompting secret sauce a) is trivial to copy, and b) gets eaten by the next model generation anyway.
I don't really understand what Cursor's business model is supposed to be long term -- at least Zed is trying to come up with new things that an editor can be (see their chat ideas and now source control). On the other hand, I also don't buy the argument some have made that Cursor and friends were banking on the marginal cost of inference going to zero, keeping prices at $20 a month, and pocketing the difference. It seems obvious (even a year ago, but moreso now) that in that situation, the IDEs also compete down to zero. If anything, higher total inference spend has to be better for them: more to skim off of. If you're already spending $50 a month on LLM tooling, Cursor doesn't have to be that big of a value-add to get you to pay them $52 instead.
I purchased Panic's Nova editor when it was first released. I'd be willing to pay Zed an equivalent ~79 per year for a perpetual license (with updates after the first year requiring further payment).
Had the same question when I read the release post. Naturally I took the ACP docs, Gemini CLI example and hacked together my own wrapper for Claude Code SDK to bridge it with Zed. It actually works and uses your Claude subscription.
https://www.npmjs.com/package/claude-code-acp
Obviously the protocol is designed to encourage other agents to build their own integration. It seems like the kind of thing LLMs would be good at, so I expect to see it soon, especially for the open source ones.
not to be confused with the other ACP from IBM, Agent Communication Protocol... Which is about "communication between agents, humans, and applications".
Threads for ongoing work, runs for stateless one-shots, long term memory. And a discovery system. No affiliation, just fun to see what folks have in their APIs.
I like Zed. It's super fast and seems to be getting lots of improvement constantly.
My issue with it is when I use it on my codebase, it doesn't like my (probably old style) eslintrc. So it decides to go ahead and reformat my file on save :(.
I've seen a lot of similar complaints. I think it would be helpful when Zed sees a new codebase to offer an onboarding questionnaire and apply the settings that can't be auto-detected from the codebase itself: coding styles, linting configuration, etc.
I would love to love Zed. In practice it’s everything I want.
But it’s a text editor first. And what I want, and it’s non negotiable, in a text editor is good text editing.
Except font rendering is atrocious and broken on Linux ( what I use ) and on Windows ( what my employer force me to use ).
For anyone using Zed, how extensive is its ability to be configured/modded to emulate other editor behaviors, bindings, etc? Eg i want something akin to Helix, but i'm not interested in Zeds Vim mode. Nor am i interested in a straight Helix mode, as my own Helix usage is quite customized.
Can i make my own Modes trees like Helix/etc offers? I'm aware of https://zed.dev/docs/helix, just not sure what the UX is to customize this behavior yourself as a user.
Readit News
I think I agree with the implicit judgment from the Zed team that it's too early to try to lock people in and capture value, and the best way to build long-term user loyalty is to just be the best editor and let people work however they want to work. On top of that, it is not a great use of their dev time to iterate on the details of an agentic coding tool when there are 10 other ones doing the same thing, and any prompting secret sauce a) is trivial to copy, and b) gets eaten by the next model generation anyway.
I don't really understand what Cursor's business model is supposed to be long term -- at least Zed is trying to come up with new things that an editor can be (see their chat ideas and now source control). On the other hand, I also don't buy the argument some have made that Cursor and friends were banking on the marginal cost of inference going to zero, keeping prices at $20 a month, and pocketing the difference. It seems obvious (even a year ago, but moreso now) that in that situation, the IDEs also compete down to zero. If anything, higher total inference spend has to be better for them: more to skim off of. If you're already spending $50 a month on LLM tooling, Cursor doesn't have to be that big of a value-add to get you to pay them $52 instead.
[0]: https://zed.dev/blog/sequoia-backs-zed#introducing-deltadb-o...
I purchased Panic's Nova editor when it was first released. I'd be willing to pay Zed an equivalent ~79 per year for a perpetual license (with updates after the first year requiring further payment).
https://github.com/zed-industries/agent-client-protocol
not to be confused with the other ACP from IBM, Agent Communication Protocol... Which is about "communication between agents, humans, and applications".
https://github.com/i-am-bee/ACP
Which is now part of A2A
https://github.com/a2aproject/A2A
I'm curious if there was an attempt to extend this to add first class support for interaction with IDEs rather than creating a new protocol.
Implemented by langchain, https://github.com/langchain-ai/agent-protocol
Threads for ongoing work, runs for stateless one-shots, long term memory. And a discovery system. No affiliation, just fun to see what folks have in their APIs.
My issue with it is when I use it on my codebase, it doesn't like my (probably old style) eslintrc. So it decides to go ahead and reformat my file on save :(.
But it’s a text editor first. And what I want, and it’s non negotiable, in a text editor is good text editing. Except font rendering is atrocious and broken on Linux ( what I use ) and on Windows ( what my employer force me to use ).
I understand it’s an alpha / beta. But still
Can i make my own Modes trees like Helix/etc offers? I'm aware of https://zed.dev/docs/helix, just not sure what the UX is to customize this behavior yourself as a user.
https://github.com/zed-industries/zed/blob/38e5c8fb66ac19f58... allows custom keybinds to actions
Modes look to be hardcoded (at the moment) https://github.com/zed-industries/zed/blob/main/crates/vim/s...