Claude Code and Claude Tag: Anthropic Redefines Engineering

A Fireside Chat with Cat and Thariq from the Claude Code Team

This month, during the AI Engineer World’s Fair, a moderator had the opportunity to lead an exciting discussion with Cat Wu and Thariq Shihipar, prominent members of the Claude Code team at Anthropic. Various topics were covered, including Claude Code, Claude Tag, Fable, the safety of coding agents, performance evaluations, as well as the design and internal use of these tools by Anthropic.

The full session is now available on YouTube. What follows is an edited version of the transcript, enriched with additional links and comments in bold.

Key Takeaways

• Claude Tag, the new collaborative integration of Slack from Claude, now accounts for 65% of the product engineering PRs for the Claude Code team.

• Claude Code first rolls out features to Anthropic employees and only launches those that show user retention within this group.

• Critical changes to Claude Code are always manually reviewed, but the team is increasingly relying on automated code review for the product’s "outer layers."

• Adding examples to a system prompt is no longer best practice for models like Fable 5 or even Opus 4.8. The system prompt for Claude Code has recently been reduced by 80%.

• Similarly, lists of "do not do X and do not do Y" can diminish the quality of results from the latest models.

• The term used at Anthropic to refer to self-use is "ant fooding."

• Anthropic firmly believes in its automatic mode, which it considers an enabling technology for Claude Tag.

• Thariq advises compensating for the undesirable effects of coding agents by being "more ambitious" in the work undertaken.

• Fable is capable of editing videos, and Thariq used it to edit his own launch video.

• Anthropic’s internal work culture in public is essential to its success, as shown by its use of Claude Tag in its public Slack channels.

How Has Daily Work Changed Over the Past Year?

Simon: Claude Code came out in February last year (it is less than a year and a half old), and it was originally just a point in the launch of Claude Sonnet 3.7. How has your daily work changed over the past year, now that we have these coding agents actually working for us?

Cat: She remembers when they first launched Claude Code and Sonnet 3.7: you had to give it a task and closely watch every little thing it tried to do. She read every permission prompt very carefully. She often said no: no, no, no, did you check this file? Did you check that other file? And now it is incredible with each model generation. She feels the team has had the chance to step back and delegate much more of the mundane implementation to Claude. That has freed up a lot of time to think about more creative work, such as: what is the right experience to offer users, now that Claude Code can implement so much of it? And now with Fable, it is a totally different step change. They see that, for many of their use cases, a great number of features can now be built with Fable.

Thariq: He remembers the first message he received about Claude Code. One of his best friends told him: "You have to try Claude Code." It was around the time Opus 4 came out, and he tried it and thought: "Oh, man. I have to work at Anthropic now." And that was Opus 4, a great model, but you had to read permission prompts. It is a bit crazy how much amnesia they have: he thinks, oh, automatic mode has always been there, right? He doesn’t even remember pressing yes and allow. For him, the big thing he is trying to focus on is that they have to produce better quality work than ever before. The results are incredibly high quality. He used it to edit videos, and he thinks, okay, this has to meet their brand team’s very strict requirements within a few hours, or they simply can’t do it. That is how he is trying to adapt with Fable: the best work they have ever done, faster than ever.

What Part of Conventional Software Engineering No Longer Holds?

Simon: What is one part of conventional software engineering that was true a year ago and that you no longer think holds in this new world?

Cat: One of the biggest changes seen in engineering skills: two years ago, it was fairly typical for a product manager to go talk to a bunch of customers, spend six months aligning with cross-functional teams on a PRD, and write a detailed specification on exactly how they were going to implement it before the first line of code was written. Now things are completely reversed. For many engineers, the encouragement she would give to the people in the room is to further develop their business sense and product sense about what they should build, because the time between having an idea and building it is so much shorter: it has gone from six to twelve months to maybe even a week. That means they all need better taste for what is worth building, which will actually influence the companies they work on. So the value of product taste and business sense is rising, and execution matters a little less in most product areas. Of course, for infrastructure, there is still a very strong emphasis on making sure every detail is right.

Thariq: For him, it is that rewrites are now good.

Simon: The worst thing you could do is now actually acceptable!

Thariq: Exactly. All the talk about the "Mythical Man-Month" (never rewrite): he is now for rewriting. If you have a good test suite (and he thinks rewriting forces you to make sure you have a good test suite), but he thinks what people underestimate is that a codebase is a specification, and perhaps it is the only copy of the specification you have, because no one knows every part of the codebase. You can take that as an artifact and distill it or create other versions. They rewrote Bun in Rust and it works great: it is live for him right now.

Simon: You are not deploying Claude Code on Bun-in-Rust yet, are you?

Thariq: Internally, they have.

(In fact, it seems Anthropic began deploying Claude Code on Bun-in-Rust for everyone on June 17.)

What Are Non-Engineers Doing with Claude Tag?

Simon: The other big launch recently was Claude Tag: it has been, what, a week now, at least for the rest of us. It is understood that it is widely used at Anthropic by non-engineers. What are non-engineers doing with Claude Tag?

Cat: Claude Tag is a Claude that lives in your team’s collaboration tools. They launched it last week in Slack. The difference with Claude Tag is that it is multiplayer by default. Once you add Claude Tag to a Slack channel, you can chime in, your teammates can chime in, and you can collaborate together on the PR. The other big difference is that it is proactive rather than reactive. You can tell Claude Tag: "Hey, monitor every bug report in this channel, set up a PR to fix it, and tag the engineer who touched that part of the codebase," and it will do so for the entire lifetime of the channel without you having to tag it manually. And the third big change is that they added team memory to it. If you tell Claude Tag your preferences in the channel, it will remember them for every future message. If you always want it to debug outages but don’t want it to debug warnings, just say so in natural language in the channel and it will remember it for you and every other member of your team.

Internally, they see Claude Tag as the evolution of Claude Code. They see it as a big change in how they work internally. Claude Tag currently accounts for 65% of their product engineering PRs.

Simon: For all of Anthropic, or just for Claude Code?

Cat: This only concerns their product engineering team: their internal version of Claude Tag accounts for 65% of their product PRs right now. And it is a huge change; it is more than 50% of their PRs. The way they see people dividing the work between Claude Code and Claude Tag is as follows: Claude Code is still the best place for your most complex tasks, when you are iterating interactively with the agent. But Claude Tag is great for working proactively on your behalf, so you no longer need to manually launch Claude Code for every bug report that comes up for the features you are working on.

Thariq: And for non-coding cases: for example, before this discussion, they asked Claude Tag: "Hey, when will Fable launch?" They wanted to make sure to coordinate it with the announcement. Claude Tag would search their Slack and see who said what. As a search engine for your company, it is really valuable. It has all the context for your product, so you can ask it questions related to metrics: often when you are making decisions, you want them to be informed by what the metrics say, so you connect it to your event store. He has seen their marketing team do things like: "Hey, tell me about this feature." They are not programmers, but Claude is a programmer: it can clone the codebase and say: "Here is the feature, here is what it looks like, here is a recording of me using the feature." This allows for a wide variety of things, and they think they are still at the beginning of discovering this.

Claude Tag as a Collaborative Team Layer

Simon: One of the problems encountered with coding agents is that it is understood how to use them as an individual, but it is not really clear how to use them in a team environment. It seems that Claude Tag is their current answer to this collaborative team layer for these tools.

Cat: Exactly. And a large percentage of their sessions are actually multiplayer right now. Maybe she says: "Hey, I think we should implement this new feature in Cowork," and she tags Claude Tag to do a first pass. Then she will tell Claude Tag: "Share a recording of your final implementation," and she will tag design to take a look. They will adjust it, then pass it to engineering to finalize and deploy it.