# Anthropic's internal dogfooding gives Claude Code rapid feedback

Markdown copy of https://realalpha.app/s/anthropics-internal-dogfooding-gives-claude-code-rapid/ on Real Alpha, the best minutes from long interviews with the people building AI. Cite and link the page, not this copy.

- Speaker: [Cat Wu](https://realalpha.app/builders/cat-wu/) · Head of product, Anthropic · Leads product for Claude Code and Cowork at Anthropic; previously a partner at Index Ventures.
- Host: [Dan Shipper](https://realalpha.app/people/dan-shipper/) · Co-founder & CEO, Every · Sold his first startup, Firefly, to Pegasystems; writes Every's Chain of Thought.
- Show: AI & I
- Episode: [The Secrets of Claude Code From the Engineers Who Built It](https://realalpha.app/v/IDSAMqip6ms/), Oct 29, 2025
- Moment: 7:57 to 14:05 of the episode (6 min), playing on https://realalpha.app/s/anthropics-internal-dogfooding-gives-claude-code-rapid/; the episode from there: https://realalpha.app/v/IDSAMqip6ms/?t=477

## Quote

> “I think **70% or 80**% of ants, technical anthropic employees, use Kali code every day.”
> — [Cat Wu](https://realalpha.app/builders/cat-wu/) at [8:24](https://realalpha.app/v/IDSAMqip6ms/?t=504) · AI & I

## Takeaways

1. 70–80% of technical Anthropic employees use it daily.
2. Internal feedback gives the team a fast signal on features.
3. Everyday workflow friction can inspire product features.

## About this segment

Cat Wu says 70–80% of technical staff use it daily, and feedback arrives every few minutes. Small team ideas, like Bash mode, then turn into product features.

From “The Secrets of Claude Code From the Engineers Who Built It”, Oct 29, 2025. [Watch the full episode](https://realalpha.app/v/IDSAMqip6ms/)

## Transcript

From the transcript AI & I publishes with the episode.

[8:01](https://realalpha.app/v/IDSAMqip6ms/?t=481) Let's bring engineers into those conversations and prioritize those surfaces and knock them out. Got it. What is AMP fooding? Oh, ant fooding is... Oh, ant fooding? Oh, it means dog fooding. Anthropic ant. I got it. Yeah, our nickname for internal employees is ant. And so ant fooding is our version of dog fooding. Internally, over... I think 70% or 80% of ants, technical anthropic employees, use Kali code every day.

[8:34](https://realalpha.app/v/IDSAMqip6ms/?t=514) And so every time we are thinking about a new feature, we push it out to people internally, and we get so much feedback. We have a feedback channel. I think we get a post every five minutes. And so you get really quick signal on whether people like it, whether it's buggy, or whether it's not good and we should unship it. You can tell. You can tell that someone that is building stuff is using it all the time to build it.

[9:02](https://realalpha.app/v/IDSAMqip6ms/?t=542) Because like it's ergonomics just makes sense if you're trying to build stuff and that that only happens if you're like ant fooding. Yeah. And I think that that's a really interesting paradigm for building new stuff like that sort of bottoms up. I make something for myself. Tell me about that. Yeah, and Cat is also so humble. I think Cat has a really big role in the product direction also. Like, it comes from everyone on the team.

[9:27](https://realalpha.app/v/IDSAMqip6ms/?t=567) And, like, these specific examples, this actually came from everyone on the team. Like, to-do lists and sub-agents, that was Sid, Hooks, Dixon shipped that, plugins, Daisy shipped that. So, like, everyone on the team, like, these ideas come from everyone. Yeah. And so I think for us, like we build this core agent loop and this kind of core experience, and then everyone on the team uses the product all the time.

[9:50](https://realalpha.app/v/IDSAMqip6ms/?t=590) And so everyone outside the team uses the product all the time. And so there's just all these chances to build things that serve these needs. Like for example, like bash mode, you know, like the exclamation mark and you can type in bash commands. This was just like many months ago, I was using quad code and I was going back and forth between two terminals and just thought it was kind of annoying. And just on the way my ass quad to kind of think of ideas, I thought of this like exclamation mark bash mode.

[10:13](https://realalpha.app/v/IDSAMqip6ms/?t=613) And then I was like, great, make it pink and then ship it. It just did it. And like that's the thing that's still kind of persisted. And, you know, now you see kind of others also kind of catching on to that. That's funny. I actually didn't know that. And that's extremely useful because I always have to open up a new tab to like run any bash commands. So you just you just do an exclamation point and then it just like runs it directly instead of filtering it through all the cloud stuff.

[10:36](https://realalpha.app/v/IDSAMqip6ms/?t=636) Yeah. And quad code sees the full output too. Interesting. That's perfect. So anything you see in the Cloud Code view, Cloud Code also sees. Okay, that's really interesting. And this is kind of a UX thing that we're thinking about. Like in the past, tools were built for engineers, but now it's equal parts engineers and model. And so as an engineer, you can see the output, but it's actually quite useful for the model also.

[10:55](https://realalpha.app/v/IDSAMqip6ms/?t=655) And this is part of the philosophy also, like everything is dual use. So for example, the model can also call slash commands. So I have a slash command for a slash commit, where I run through kind of a few different steps, like diffing and generating a reasonable commit message and this kind of stuff. I run it manually, but also Claude can run this for me. And this is pretty useful because we get to share this logic.

[11:17](https://realalpha.app/v/IDSAMqip6ms/?t=677) We get to kind of define this tool, and then we both get to use it. Yeah. What are the differences in designing tools that are dual use from designing tools that are used by one or the other? Surprisingly, it's the same. Okay. So far. Yeah. I sort of feel like this kind of elegant design for humans translates really well to the models. So you're just thinking about what would make sense to you and the model generally, it makes sense to the model too, if it makes sense to you.

[11:47](https://realalpha.app/v/IDSAMqip6ms/?t=707) Yeah, I think one of the really cool things about Cloud Code being a terminal UI and what made it work really well is that Cloud Code has access to everything that an engineer does at the terminal. And I think when it comes to whether the tool should be dual use or not, I think making them dual use actually makes the tools a lot easier to understand. It just means that, okay, everything you can do, Cloud Code can do, there's nothing in between.

[12:13](https://realalpha.app/v/IDSAMqip6ms/?t=733) That's interesting. Yeah, there are a couple of those decisions. So no code editor. It's in the terminal so it has access to your files. And it's on your computer versus like in the cloud in a virtual machine. So you get like repeated, you get to use it in a repeated way where you can like, you know, build up your Cloud MD file or, you know, like all like build slash commands and all that kind of stuff where it becomes very composable um and extensible from a very simple starting point and i'm curious about how you think about you know for for people who are thinking about okay i want to build an agent i want to build probably not cloud code but like something else how you get that that simple package that then can extend and be really powerful over time hmm For me, I start by just thinking about it like developing any kind of product where you have to solve the problem for yourself before you can solve it for others.

[13:13](https://realalpha.app/v/IDSAMqip6ms/?t=793) And like this is something that they teach in YC is you have to start with yourself. So like if you can solve your own problem, it's much more likely you're solving the problem for others. And I think for coding, starting locally is the reasonable thing. And now we have quad code on the web. So you can also use it with a virtual machine. And you can use it in a remote setting. And this is super useful when you're on the go.

[13:34](https://realalpha.app/v/IDSAMqip6ms/?t=814) You want to take something off from your phone. And this is sort of, we started proving this out kind of a step at a time where you can do at quad in GitHub. And I use this every day. Like on the way to work, I'm like at a red, white, I probably shouldn't be doing this. But I'm like, on GitHub, at a red, white. And then I'm like, add Claude, fix this issue or whatever. And so it's just really useful to be able to control it from your phone.

[13:56](https://realalpha.app/v/IDSAMqip6ms/?t=836) And this kind of proves out this experience. I don't know if this necessarily makes sense for every kind of use case. For coding, I think starting local is right.