00:03Let's get started

00:06Welcome everyone to Cloud Code best practices

00:09In this talk, I'm going to talk about kind of what cloud code is at a high level

00:14Then we'll peer under the hood a little bit to kind of understand how cloud code works, and then knowing that, because it's useful to kind of know how your tools work, we're going to talk about good use cases for cloud code and also best practices we've figured out both internally and from our users, uh, for getting the most out of this tool

00:35Uh, but before I get started, I'd like to introduce myself real quick and talk about how I ended up on the stage

00:41So, my name's Cal and I joined Anthropic about a year and a half ago, uh, to help start up a team we call Applied AI

00:48And it's the Applied AI's kind of mission, our team's mission, is to help our customers and partners build great products and features on top of Claude

01:00So what that really means is I spend a lot of my day prompting Claude to get the absolute best outputs out of these models

01:07That said, I also love to code, and I'm definitely one of those coders that like starts a lot of projects, has some crazy idea, and then just never finishes them

01:16So, I have this graveyard of just like code that I started, never really

01:19Finished

01:23Um, but I'm always spinning new things up, and late last year, I was in Slack, and I was hearing about this new tool that a few people are using

01:27They were saying it was really cool, and so, on a Friday night, I downloaded the tool that would become Claude Code, and I threw it at this kind of new notetaking app that I wanted to build

01:36And like that whole weekend just kind of totally changed the way that I code and think about software engineering

01:42I was carrying around my laptop with me all weekend

01:44I was super addicted to just watching Claude Code work, and I would press enter, and I'd switch over to my browser and refresh, and I watched this huge, powerful application come together in front of my eyes, and I got way farther into this thing than I ever would have on my own, and it just blew my mind

02:01And while I was doing this, I was a little worried

02:05I was like, you know, I, you know, I kind of know how these things work, so I'm like, man, I'm using a lot of tokens

02:08I hope I don't get in trouble or anyone like notices

02:12I'm not really contributing to Anthropic code

02:13Um, but what I didn't know is that the Claude code team had built this internal, like, leaderboard tracking how much all the Anthropic employees were

02:23Using this, and over the weekend, I had shot to the top

02:27And so through that, I got to meet Boris and Cat and some of the early Claude Code team

02:31And I was able to start talking to them and say, "Hey, I love this tool

02:34I also know a lot about prompting

02:36Can I help you all out?" And so through that, I got involved, and now I'm one of the core contributors on the team, and I do a lot of, I work a lot on the prompting, the system prompts, how the tools work, the tool descriptions, and tool results, as well as I work on how we evaluate this tool

02:54So when we think about changing the prompts, how do we make, how do we know we made things better? or the same, and we didn't totally ruin Claude Code

03:02So with that said, let's kind of dive in

03:05So, here's my current mental model of Claude Code and how I describe it to people when people ask me

03:12Claude Code is like that co-worker that does everything on the terminal

03:15It's the sort of person that just never touches the GUI

03:22They're a whiz

03:29I think of when I was a junior engineer, I had this mentor, and I would walk over to his desk, and I would say, "Hey, Tony, can you help me with this bug?" and he would whipping it open his terminal, and he'd be like doing all these crazy bash commands and changing things around in Vim, and I'd always walk away thinking, "Wow, that was crazy

03:36I should learn how to do that." Um, I never did, but having Claude Code on your computer is kind of like having Tony next to you all the time

03:49So, how does Claude code kind of work under the hood? At Anthropic, we try to always do what we call the simple thing that works, and what that means for Cloud Code is it's what we would consider a very pure agent

04:03And Anthropic, when we talk about agents, what we really mean is some instructions, some powerful tools, and you let the model just run in a loop until it decides it's done, and that's really what Cloud Code is, so it's tools, powerful tools, and the tools that you know someone that was really good at a terminal would be able to use, tools to create and edit files, to use the terminal

04:29And then you can also do things like pull in other things with MCP

04:35Now, on top of that, there's how Claude understands the codebase

04:39And if you're going to build a coding agent or a coding tool a year ago, you'd probably have ideas like, well, okay, I'm going to get this user message about something about this codebase and I'll need to figure out which files are relevant, so maybe I'll like index the whole codebase and

04:55Embed it and do this fancy, like, kind of RAG retrieval thing

04:58That is not how Claude code works

05:01We don't do any sort of indexing

05:03Instead, Claude kind of explores and understands the codebase, how you, if you were new to a team and new to a codebase, would explore a codebase, and that is through a genetic search, is the same sort of search tools you or I would use, things like glob and grep and find, and it can work its way through a codebase and understand what's going on, and when we talk about a genetic search, that really means the model can go do some searches, and then it can look at the results and can say, "Hmm, maybe I need to..." Figure out a few more things

05:32I'm going to go do some more searching and then come back, and then on top of these primitives, on top of this agent, we have a few things

05:41We have a very nice light UI layer where you get to watch Claude code work

05:44You see all the text fly by, and we have this nice permission system that allows the agent to work and allows, and kind of forces the human to butt in when the agent is doing something dangerous

05:55And then on top of that, we also care a lot about security in this tool

06:00And so because quad code is just such a lightweight kind of layer on top of the model, and the fact that our model is Available not just behind Anthropic APIs, but also with our cloud providers, AWS and GCP

06:15It's very easy and native to point cloud code at one of these other services if you feel more comfortable consuming cloud that way

06:26Now, a lot of people ask me, "Hey Cal, what can I use cloud code for? Like, what is it good at? Where is it interesting?" And the reality is, it's kind of great at everything

06:34So let's start with discovery

06:42Often times in your career, you will be dropped into a new codebase, whether that means you're switching teams, you're switching companies, I don't know, you're starting to work on some sort of

06:48open source project

06:53And probably when you're first getting started and getting familiar, you're not very productive because you're just trying to figure out where things are in the codebase, what patterns kind of the team is using, things like that

06:59And Claude Code can kind of help supercharge that onboarding process

07:04You can ask Claude, "Hey, where is this feature implemented?" Or since it's great at the terminal, you can say, "Hey, look at this file and look at the git history and just kind of tell me a story about how this code has changed over the past couple weeks." One thing you can use Claude Code For, and I think this is underrated, is instead of just diving in and starting to work, you can use Claude Code as a thought partner

07:27So often times when I'm working with Claude and I want to implement a feature, or we're going to change something up, I'll open up Claude and I'll say, "Hey Claude, you know, I'm thinking about implementing this feature, can you just kind of like search around and kind of figure out how we would do it and maybe report back with like two or three different options

07:44Don't start working, don't start writing any files, writing any files yet," and Claude will go off and use those

07:50Agentic search capabilities and come back with a few ideas, and then I could work with Claude to kind of validate things, and then we can jump into the project

08:01Of course, Cloud Code is great at building and writing code, and I would say this in on two different fronts

08:04One, it can do the zero to one sort of stuff

08:09You drop it in an empty directory and you say, "Hey, build me an app, build me a game," that demos where very well, it's very fun to do, it's very gratifying

08:16Of course, in reality, what really matters is, is Cloud Code good working in existing code bases? And this is primarily what we focus on, um, on the Cloud Code

08:28Team, we have in our codebase abnormally high, I would say, unit test coverage, and that's because Claude code makes it so easy and just straightforward to add unit tests

08:39So, we have great code coverage

08:41And then the other thing we have in Claude code in our own codebase is we have great commits and PR messages, because when we finish working, we'll just say, "Hey Claude, write the commit for me, write the PR message for me." We also see great opportunities to use Claude code in kind of the deployment, like deployments, and in other parts of the life cycle, and this is a few other people have talked about this, but this

09:05Is using the Claude Code SDK, so using it headlessly, using it programmatically, being able to sprinkle in a coding agent anywhere

09:14And so that's things like sprinkling it into CI/CD to use it in GitHub, for instance, to help people um programmatically

09:24And then finally, it's great kind of with support and scale

09:26It can help you debug errors faster

09:28One thing that we saw when we started giving Claude Code to customers and talking to them about it, we didn't totally predict this was a lot of customers or potential customers said, "Hey, we've been, we've been kind of putting off this like large Codebase migration, people that are on old versions of Java trying to get to a new one, or a team that's on PHP and they're trying to get to React or Angular

09:51We've talked to multiple teams like this, and having a tool like Cloud Code makes projects like that a little more digestible

09:57When you go to your team and you say, "Hey, we're going to spend a month, you know, refactoring or rewriting large parts of the codebase." And then on top of that, and this kind of matters across all these, is once again, remember Claude is great at the terminal, and that means it's going to be great at all those different CLI

10:16Tools, things like Git, Docker, Big Query, things like that

10:18I never have to worry about, oh, I'm going to get myself, how do I get myself out of this sticky rebase? I'll just fire up cloud code and tell it the situation and be like, "Hey, can you fix this for me?" It's incredible

10:33Now, let's talk about best practices, and the first one is not going to be a surprise, but the first one is use claw.md files

10:41So remember that cloud code, like I said, is an agent and it has some tools, has some lightweight instructions in the prompt, but it doesn't really have memory

10:51And so the main way we share state across kind of sessions or across our team when we fire up Claude code in the same codebase over and over again, is this claude.md file

11:04So when we start Claude, what happens is, if there's this claude.md file in the working directory, it's just plopped into context

11:11It's plopped into the prompt, and basically what it says is, "Hey Claude, by the way, these are important instructions the developer left for you

11:18Be sure to pay close attention to this." And there's various places you can put the claude.md file

11:23You can put it in a project and check it in so all your teammates share it

11:27You could put one in your home directory if there's Things you just want Claude to always know about, regardless of what you're working on

11:33And the things you put in here are things like, hey, by the way, maybe this is how you run the unit tests

11:38Or just so you know, to make kind of your searching and life easier, here's like just like an overview of kind of how this project is laid out, where the tests live, what different modules are, things like that, or here's our style guide

11:52All sorts of things like that to just make Claude's life a bit easier, and you can build these things up over time

12:01The other thing you can do, which is important, is permission

12:02management

12:05When you're running Claude Code, there's all sorts of different kind of permission things flying by

12:08Kind of out of the box, what happens when you start our tool is for read actions

12:12If Claude is searching or reading, we just let it go, but once it starts writing or running bash commands or doing things that could change stuff on your machine potentially, that's when we kick in this UI and it says something like, yes, yes, always allow this or no, um, I want to do something else

12:33And using that permission management and being smart about it can help you work faster

12:40So there's Something called auto-accept mode, where if you're working with Claude Code and you press shift tab, Claude will just start working

12:46There's things you can do, like you can configure Claude in the settings where specific commands, like on bash, like if you just are like tired of saying yes, run npm run test, you can just always approve that

12:58So, fiddling with your permission management is a great way to kind of speed up your workflow integration setup

13:06So, one thing that is going to help you get the most out of Claude Code is remember that it's great at the terminal, and if there's applications that you use which Have kind of a way to access them through CLI, and GitHub is a great example of that

13:14They have a powerful tool called GH

13:23You can basically give more work to Claude code, and you can do that either by just installing more CLI tools, or you can attach more MCP servers

13:31Um, I would say, just through experience, that if you're using something like, um, a CLI tool that's well known and well documented, and you're trying to choose between the CLI tool and just installing it on your machine and grabbing an MCP server, I would recommend using the CLI tool

13:49Um, and then also, if you internally have your own tools at Anthropic, we have something called Coup that does a whole bunch of stuff for us

13:58You can also tell Claude about that, and you, that's the sort of thing you'd put in claude.mmd, and then context management

14:06So remember that Claude is an agent, and when it's an, what, what it does, it's calls these tools, and the context builds up and up over time, and at least for Anthropic, our models have a context window of 200,000 tokens, and you can max this thing out

14:23So you kind of have two options when you're in a long session with Claude, and you're working, and you're going back and forth

14:28You'll see in the bottom right

14:30You'll start to get this little warning that'll say, "Hey, you're starting to fill up the context window," and kind of depending on what's going on, you have two options

14:37You can run slash clear and just start over, and that clears everything out except for, for instance, claw.mmd, or you can run slash compact, and what'll happen is basically, it's like a user message is inserted, and it just says something like, "Hey, I need to go summarize everything we've been up to

14:55I'm going to give this to another developer, and they're going to pick up where I left off." And then that summary is what kind of seeds the next session

15:01Can go from there

15:08We spend a lot of time tuning this kind of compact functionality so that as you max out the context window and then run compact, you can start back over and keep going, efficient workflows

15:14What can you do with Claude Code? And how do you get the most out? So using planning and to-dos, I talked a little bit about this before, but one of the best things you can do is when you open up Claude Code, instead of saying, "Hey, I need you to fix this bug," you can say, "Hey, I have this bug

15:30Can you search around, figure out what's causing it, and just like tell me a plan how we're going to fix it?" And this can

15:36Save you a lot of time because you can verify, you can read Claude's plan, and you can verify what it's going to do

15:42And then the other thing that we have is we have this to-do list feature

15:46So often when Claude's working on a big task, it'll create a to-do list, and if you're kind of paying attention, you can kind of watch this to-do list, and if you see anything kind of weirder in there or something that doesn't make sense, that's when you can press escape and say, "Hey Claude, let's change the to-do list

16:01I think you're on the wrong path." Smart vibe coding

16:03So it's very tempting and it's very powerful to Just let Claude work and press enter and see what happens at the end

16:10I think there's a few things that can help make this better, and there's, I think, a talk later today about just this for 30 minutes

16:19But doing things like having test-driven development, having Claude make small changes, run the tests, make sure they pass, always having Claude do things like check the TypeScript and the linting, and then commit regularly so that if it's kind of going off the rails, you can always fall back and try again

16:35You can use screenshots to guide and debug, so Claude is built on top of our models which are multimodal

16:41You can always just grab a screenshot, paste it in, or if you have a file somewhere that's an image, you can just say, "Hey, Claude, look at this mock.png and then build the website for me or whatever." And then advanced techniques

16:55So, as you're getting used to using Claude, what are some things you can think about, uh, to kind of push things to the next level? And one of the things we see both internally and with customers is, when you've started to use this tool for a while, it's going to be very tempting to use multiple Claudes at once

17:12And so I know people at Anthropic and a few customers that run four Claudes at the same time

17:18There's various ways to do this

17:20You can have it in T-Mox or just different tabs, all sorts of crazy things

17:25So I would challenge you to try getting multiple Claudes running at once and kind of be orchestrating all these things

17:28It's quite fun

17:31I can only do two, but I know people that do four

17:33Use escape

17:37So, escape is your best friend

17:40While Claude is working, you can kind of keep an eye on what it's up to, and you can press escape to stop it and interject and say, "Hey, I think you're going on the wrong path, or I want you to do something else." Knowing when the right time to press

17:52Escape, versus just letting Claude figure it out, is key to getting the most out of the tool

17:56And there's a hidden feature

18:00Not too many people know about it, but if you press escape twice, you can actually jump back in your conversation

18:03You can go back and you can kind of reset tool expansion in MCP

18:09So, this is taking it to the next level

18:12If you feel like with bash and with the tools that Claude has, that it still can't do something, this is when you should start looking at MCP servers

18:18And then headless automation

18:25And I think this is the thing we're most excited about, but also we are still trying to wrap Our heads around internally, which is how can we use Claude programmatically

18:31We have that in GitHub actions

18:34We want to figure out other creative places we can start using it

18:38I would challenge you all to do the same

18:43So, with that said, uh, I'm going to jump over to my computer because there's one other best practice, which is it's always good to stay on top of everything that's new

18:53So, we're shipping super fast

18:55I'm going to throw, I'm just going to go over a few things that are new as of today

19:01Um, one thing is when you're in Cloud Now and you fire it up, you can do slashmodel

19:04You can see what model you're

19:07Running on, I'm on default, which happens to be Sonnet

19:09We can jump over to Opus

19:11You can do the same thing in slash config, switch it here, so that's new

19:18Make sure you're running the model that works for you

19:20There's another thing that's new about these models, which is you can say something like, "Um, can you figure out what's in this project?" And for a long time, for a while, we've had this like, "think hard" or "extended thinking." Now this is great, but with our past models, we wouldn't let our model think between tool calls, and that's probably when the thinking matters most

19:44So starting with Claude 4, they can

19:46Now our models now think between tool calls, and we can watch this happen

19:50So we have Claude in this project

19:51There's a few different files in here, and I'm just going to tell it to think hard and figure out what's in this project, and we can watch Claude start to work

19:59And so the way you know you triggered thinking is you'll see kind of this lighter gray text, and then it'll call some file, it'll call some tools, it'll read some stuff, and then we see some more thinking

20:09And this is awesome

20:13Um, so I encourage you when you're working on tasks and solving bugs, throw a "think hard" in there

20:19And then the other thing, And you know what, we'll just throw it up real quick, is I have this in VS Code, but of course this is in Jet Brains as well, but we have these new great integrations with VS Code and Jet Brains

20:31Um, we can do things like Claude's going to know what file I'm in

20:36What file am I in? That is not what I meant to say, but Claude's going to figure it out, and you can do things like this

20:58So these are the sort of things I would encourage you to stay on top of

21:00We have a public, uh, kind of GitHub project called Claude Code under Enthropic

21:05You can post issues there, but we also post our change log there, and so I check this once

21:12A week and make sure that I'm on top of all the new stuff we're shipping, because even I can't keep up with it

21:16So, with that said, we have like four minutes left

21:19I'm happy to answer questions about anything Claude code related

21:21We have it here

21:24I can live demo some stuff if you're interested

21:26Um, let's do a few

21:32Thanks

21:39Real quick, this might be obvious, but multiple cloud MD files in a project, I presume that's possible and it just figures it out or no? So, there's a few options, of course, like in the same directory

21:45You couldn't, um, but you could have one here and one in a subdirectory, and I think we changed

21:56This, so that all the subdirectory ones aren't read in, because like Anthropic, we have a monorepo and people would open it at the top and blow up their context with all the Claude MDs

22:07So, we encourage Claude, when it's searching around and it discovers claw.md files in child directories that are relevant, to be sure to read them, but by default, it just reads the cloud MD file in the current working directory when you fire it up, and then also you can set one in like your home directory

22:25Um, there are things you can do though

22:28We have this new thing, like in your cloud MD, you can start referencing other files

22:35So you could, for instance, um, do something like this with an @ sign, um, if you have other Claude MD files that you just kind of know you always want to read in, um, to do something like that

22:39Hi

22:43Okay

22:48I, um, have not had luck getting Claude to respect my Claude MD

22:52Like there's one thing in particular, yes, where I'll ask it to refactor something, and then it will leave inline comments explaining the, like, the what of it is, and it's like, like something that's extremely obvious, and so I'll tell it, like, go and remove any inline comments that describe the what of what's happening, and then it will remove it, and then

23:11Immediately do it again and like the same pass, so do you have any strategies for dealing with that? So there's kind of two things that fix that

23:18So that was actually kind of a model problem

23:20There's nothing in the prompt

23:21We have actually a lot in the prompt for 37 that said, "Whoa, do not leave comments," and despite that, the model just loves to leave comments

23:28Um, so it doesn't surprise me that your Cloud MD didn't help much either

23:31We already did a lot, I did a lot of work to try to tamp it down from what happens out of the box, so we mostly fixed that in Cloud 4

23:41Now there might be some new weird behavior

23:44Quirks, but the other thing we made better in Claude 4 is it's just better at following instructions

23:49Um, and we've gotten a lot of feedback from early testers that, uh, all of a sudden, whoa, my Claude MD is being followed way more closely

23:56Um, and it might be a good chance to go look in your ClaudeMD and decide, do I still need this stuff? Maybe I can take some of it out

24:04Maybe I need to add a few new things

24:06So, moving over to the new models might be a good time to take another look at what's in there and see what you need and what maybe can go

24:14Uh, for the record, I'm trying to think of something that you might not have

24:16thought of

24:18We're doing multi-agent execution and parallelization

24:20Can you make it so that for four agents, say agents two and three, use the context from agent one, maybe agent four uses the context from agent two at a certain point? Yeah

24:32Um, yeah, etc

24:35That's interesting

24:38We're trying to—so, kind of like I said at the beginning, we're trying to do the simple thing that works, which is just one agent that's great at coding and does everything

24:41Um, I think we want to figure that out

24:46Probably what's going to happen is, if you wanted to do that, you would ask all your agents to probably like write to a shared Markdown file or something like that, so they can all kind of, like, check in and communicate

24:56Um, sometimes, like, I'll be working with cloud.md or Claude, and I'll just say, like, "Hey, I need you to write some stuff in, like, ticket.md for another developer," and then I'll fire up another Claude code, and I'll be like, "Hey, read ticket.md, like another developer left this note for you

25:10Like, this is what you're going to work on." So, I would think about trying to write that state to a file and then just kind of, like, count on the model's ability to just, like, read files and make sense of them, um, is probably the best you can do

25:24Today, and maybe we'll figure out clever ways to expose that, uh, in the product as something more native

25:32Cool

25:36All right

25:39And with that said, I have some rare Claude code stickers that I found in my backpack, so come find me

25:42I'll be hanging out over there or something

25:43Um, happy to share them

25:46Thank you

25:47[Applause]