[
  {
    "i": 0,
    "speaker": "Speaker 1",
    "text": "Welcome to Code with Claude, everyone. I'm Jess, product for Claude Managed Agents."
  },
  {
    "i": 1,
    "speaker": "Speaker 2",
    "text": "And I'm Michael, engineering for Claude Managed Agents."
  },
  {
    "i": 2,
    "speaker": "Speaker 1",
    "text": "It has been so exciting to see the organic uptake on Claude Managed Agents and how we're accelerating meaningfully developer workflows."
  },
  {
    "i": 3,
    "speaker": "Speaker 1",
    "text": "This is a story that we want to share with you today."
  },
  {
    "i": 4,
    "speaker": "Speaker 1",
    "text": "We've seen uptake from disruptive startups all the way to the largest enterprises, and we want to make sure that you can all benefit from the kind of experiences that they've had as well."
  },
  {
    "i": 5,
    "speaker": "Speaker 1",
    "text": "So first, we'll go over the exponential that we've seen in AI capabilities and what that implies for building agents."
  },
  {
    "i": 6,
    "speaker": "Speaker 1",
    "text": "We'll talk through the patterns we saw in agentic development and why that motivated us to build Claude Managed Agents."
  },
  {
    "i": 7,
    "speaker": "Speaker 1",
    "text": "We'll go through the building blocks for building an agent, including some of the most important ones."
  },
  {
    "i": 8,
    "speaker": "Speaker 1",
    "text": "We'll go over some of our more recently announced features."
  },
  {
    "i": 9,
    "speaker": "Speaker 1",
    "text": "And we'll round things out with a fireside chat with one of our hero Japanese users, Rakuten."
  },
  {
    "i": 10,
    "speaker": "Speaker 1",
    "text": "So as we've all seen, models are getting exponentially more capable, and so are our expectations."
  },
  {
    "i": 11,
    "speaker": "Speaker 1",
    "text": "The more advanced our model capabilities, the more sophisticated the work that we're delegating."
  },
  {
    "i": 12,
    "speaker": "Speaker 1",
    "text": "We're now seeing that the bottleneck is increasingly the infrastructure and not actually the intelligence."
  },
  {
    "i": 13,
    "speaker": "Speaker 1",
    "text": "So let's bring this to life."
  },
  {
    "i": 14,
    "speaker": "Speaker 1",
    "text": "A few years ago, when Opus 3 was announced, you might have it write and test a single component."
  },
  {
    "i": 15,
    "speaker": "Speaker 1",
    "text": "It would take minutes of focused work."
  },
  {
    "i": 16,
    "speaker": "Speaker 1",
    "text": "Last year, when our Claude 4 models came out, we leveled up, and you could debug an entire set of files."
  },
  {
    "i": 17,
    "speaker": "Speaker 1",
    "text": "You might be working for an hour or so, but you'd be steering heavily along the way."
  },
  {
    "i": 18,
    "speaker": "Speaker 1",
    "text": "With our latest models this year, you're now seeing them run overnight across agent teams, listening to your linear backlog, and actioning the whole thing before you wake up."
  },
  {
    "i": 19,
    "speaker": "Speaker 1",
    "text": "And now we suspect that in the near-term future, with models as capable as Fable, we'll enter a world where agents are capable enough to accomplish tasks that would have previously taken entire teams of people quarters to achieve."
  },
  {
    "i": 20,
    "speaker": "Speaker 1",
    "text": "And agents will run this entirely autonomously."
  },
  {
    "i": 21,
    "speaker": "Speaker 1",
    "text": "So you might see that multi-agent systems could coordinate and run an entire M&A pipeline, end-to-end, in a fraction of the time it would take us."
  },
  {
    "i": 22,
    "speaker": "Speaker 1",
    "text": "And so, as tasks have graduated"
  },
  {
    "i": 23,
    "speaker": "Speaker 1",
    "text": "from low-level instructions towards end-to-end outcome description."
  },
  {
    "i": 24,
    "speaker": "Speaker 1",
    "text": "We now need far more than just prompts and tool loop."
  },
  {
    "i": 25,
    "speaker": "Speaker 1",
    "text": "We need reliable and scalable agentic infrastructure."
  },
  {
    "i": 26,
    "speaker": "Speaker 2",
    "text": "That's exactly right."
  },
  {
    "i": 27,
    "speaker": "Speaker 2",
    "text": "And the more complex these tasks,"
  },
  {
    "i": 28,
    "speaker": "Speaker 2",
    "text": "the deeper the access we'll need to give these agents"
  },
  {
    "i": 29,
    "speaker": "Speaker 2",
    "text": "in order for them to be effective."
  },
  {
    "i": 30,
    "speaker": "Speaker 2",
    "text": "With models like Fable, we really need these to..."
  },
  {
    "i": 31,
    "speaker": "Speaker 2",
    "text": "They're able to do very complex tasks."
  },
  {
    "i": 32,
    "speaker": "Speaker 2",
    "text": "You can't run an effective agent"
  },
  {
    "i": 33,
    "speaker": "Speaker 2",
    "text": "without granting it access to your credentials,"
  },
  {
    "i": 34,
    "speaker": "Speaker 2",
    "text": "your internal knowledge bases, or your databases."
  },
  {
    "i": 35,
    "speaker": "Speaker 2",
    "text": "If you want these agents to produce code for you,"
  },
  {
    "i": 36,
    "speaker": "Speaker 2",
    "text": "you need to give them access to your actual code bases"
  },
  {
    "i": 37,
    "speaker": "Speaker 2",
    "text": "so that they can push up PRs and go to production."
  },
  {
    "i": 38,
    "speaker": "Speaker 2",
    "text": "And finally, you need to grant them identity and authentication."
  },
  {
    "i": 39,
    "speaker": "Speaker 2",
    "text": "Our agents are increasingly acting not just as Claude,"
  },
  {
    "i": 40,
    "speaker": "Speaker 2",
    "text": "but as me or Jess, with our emails and our Slack."
  },
  {
    "i": 41,
    "speaker": "Speaker 2",
    "text": "And as we grant agents these human-like capabilities,"
  },
  {
    "i": 42,
    "speaker": "Speaker 2",
    "text": "we expect them to react."
  },
  {
    "i": 43,
    "speaker": "Speaker 2",
    "text": "We expect them to interact with them in much more human-like ways."
  },
  {
    "i": 44,
    "speaker": "Speaker 2",
    "text": "The shape of interactions are changing,"
  },
  {
    "i": 45,
    "speaker": "Speaker 2",
    "text": "not just the duration of these interactions."
  },
  {
    "i": 46,
    "speaker": "Speaker 2",
    "text": "Some agents are very conversational."
  },
  {
    "i": 47,
    "speaker": "Speaker 2",
    "text": "You steer, provide guidance along the way."
  },
  {
    "i": 48,
    "speaker": "Speaker 2",
    "text": "You might even interrupt it if you think that it's going off path."
  },
  {
    "i": 49,
    "speaker": "Speaker 2",
    "text": "Some agents on new models like Fable"
  },
  {
    "i": 50,
    "speaker": "Speaker 2",
    "text": "are really, really outcome-oriented."
  },
  {
    "i": 51,
    "speaker": "Speaker 2",
    "text": "And if you have strong signal or a rubric"
  },
  {
    "i": 52,
    "speaker": "Speaker 2",
    "text": "of what you want to see get done,"
  },
  {
    "i": 53,
    "speaker": "Speaker 2",
    "text": "you can provide it that rubric and let it iterate"
  },
  {
    "i": 54,
    "speaker": "Speaker 2",
    "text": "until your exit criteria are met."
  },
  {
    "i": 55,
    "speaker": "Speaker 2",
    "text": "And then finally, you might start a task a couple days ago"
  },
  {
    "i": 56,
    "speaker": "Speaker 2",
    "text": "and then want to pick it up much, much later."
  },
  {
    "i": 57,
    "speaker": "Speaker 2",
    "text": "A robust agent platform needs to support"
  },
  {
    "i": 58,
    "speaker": "Speaker 2",
    "text": "all these sorts of interaction patterns."
  },
  {
    "i": 59,
    "speaker": "Speaker 2",
    "text": "And the infrastructure and primitives that we provide"
  },
  {
    "i": 60,
    "speaker": "Speaker 2",
    "text": "have to give you all of this out of the box"
  },
  {
    "i": 61,
    "speaker": "Speaker 2",
    "text": "while still being very, very flexible"
  },
  {
    "i": 62,
    "speaker": "Speaker 2",
    "text": "so that you can customize them to your needs."
  },
  {
    "i": 63,
    "speaker": "Speaker 1",
    "text": "So it's now becoming clear that we expect a lot of our agents."
  },
  {
    "i": 64,
    "speaker": "Speaker 1",
    "text": "And historically, this has meant"
  },
  {
    "i": 65,
    "speaker": "Speaker 1",
    "text": "that we've pushed the burden onto you, the developer."
  },
  {
    "i": 66,
    "speaker": "Speaker 1",
    "text": "In research, we conducted,"
  },
  {
    "i": 67,
    "speaker": "Speaker 1",
    "text": "prior to launching Cloud Managed Agents,"
  },
  {
    "i": 68,
    "speaker": "Speaker 1",
    "text": "we saw that developers were genuinely eager"
  },
  {
    "i": 69,
    "speaker": "Speaker 1",
    "text": "to climb the exponential with us,"
  },
  {
    "i": 70,
    "speaker": "Speaker 1",
    "text": "but were struggling with a few key areas."
  },
  {
    "i": 71,
    "speaker": "Speaker 1",
    "text": "So first, context management."
  },
  {
    "i": 72,
    "speaker": "Speaker 1",
    "text": "The right context at the right time"
  },
  {
    "i": 73,
    "speaker": "Speaker 1",
    "text": "is actually really difficult to tune"
  },
  {
    "i": 74,
    "speaker": "Speaker 1",
    "text": "even though it's completely necessary."
  },
  {
    "i": 75,
    "speaker": "Speaker 1",
    "text": "And providing the context at the wrong time"
  },
  {
    "i": 76,
    "speaker": "Speaker 1",
    "text": "can be a huge distraction for your agents."
  },
  {
    "i": 77,
    "speaker": "Speaker 1",
    "text": "Half of our developers are citing"
  },
  {
    "i": 78,
    "speaker": "Speaker 1",
    "text": "that infrastructure concerns"
  },
  {
    "i": 79,
    "speaker": "Speaker 1",
    "text": "are their number one production blocker."
  },
  {
    "i": 80,
    "speaker": "Speaker 1",
    "text": "So agents create bursty workloads,"
  },
  {
    "i": 81,
    "speaker": "Speaker 1",
    "text": "they have unpredictable compute patterns."
  },
  {
    "i": 82,
    "speaker": "Speaker 1",
    "text": "It's super difficult to scale securely"
  },
  {
    "i": 83,
    "speaker": "Speaker 1",
    "text": "while also hitting latency targets."
  },
  {
    "i": 84,
    "speaker": "Speaker 1",
    "text": "And lastly, observability is genuinely incredibly difficult."
  },
  {
    "i": 85,
    "speaker": "Speaker 1",
    "text": "How do you know when your agent"
  },
  {
    "i": 86,
    "speaker": "Speaker 1",
    "text": "is producing quality outputs?"
  },
  {
    "i": 87,
    "speaker": "Speaker 1",
    "text": "These are non-deterministic models"
  },
  {
    "i": 88,
    "speaker": "Speaker 1",
    "text": "and they're producing huge amounts of unstructured data."
  },
  {
    "i": 89,
    "speaker": "Speaker 1",
    "text": "And so enter Cloud Managed Agents."
  },
  {
    "i": 90,
    "speaker": "Speaker 1",
    "text": "We did the platform work so that you don't have to."
  },
  {
    "i": 91,
    "speaker": "Speaker 1",
    "text": "Managed Agents combines infrastructure, agentic primitives,"
  },
  {
    "i": 92,
    "speaker": "Speaker 1",
    "text": "and the ability to manage your data."
  },
  {
    "i": 93,
    "speaker": "Speaker 1",
    "text": "And so enter Cloud Managed Agents."
  },
  {
    "i": 94,
    "speaker": "Speaker 1",
    "text": "We did the platform work so that you don't have to."
  },
  {
    "i": 95,
    "speaker": "Speaker 1",
    "text": "Managed Agents combines infrastructure, agentic primitives,"
  },
  {
    "i": 96,
    "speaker": "Speaker 1",
    "text": "and out-of-the-box observability,"
  },
  {
    "i": 97,
    "speaker": "Speaker 1",
    "text": "all available in a package on the Cloud platform."
  },
  {
    "i": 98,
    "speaker": "Speaker 1",
    "text": "We'll dive into each of these components in more detail"
  },
  {
    "i": 99,
    "speaker": "Speaker 1",
    "text": "throughout the course of this presentation"
  },
  {
    "i": 100,
    "speaker": "Speaker 1",
    "text": "along with a couple of demos."
  },
  {
    "i": 101,
    "speaker": "Speaker 2",
    "text": "So let's talk about the very basic building blocks"
  },
  {
    "i": 102,
    "speaker": "Speaker 2",
    "text": "of Cloud Managed Agents."
  },
  {
    "i": 103,
    "speaker": "Speaker 2",
    "text": "At the very, very core, you have an agent that you define."
  },
  {
    "i": 104,
    "speaker": "Speaker 2",
    "text": "This would be the system prompt, the model you want to use,"
  },
  {
    "i": 105,
    "speaker": "Speaker 2",
    "text": "any skills that you might want to have loaded"
  },
  {
    "i": 106,
    "speaker": "Speaker 2",
    "text": "into your agent, and tools with permissions"
  },
  {
    "i": 107,
    "speaker": "Speaker 2",
    "text": "for those tools that you want that agent to have."
  },
  {
    "i": 108,
    "speaker": "Speaker 2",
    "text": "This is like the agent's identity."
  },
  {
    "i": 109,
    "speaker": "Speaker 2",
    "text": "Next, you have the environment that you configure."
  },
  {
    "i": 110,
    "speaker": "Speaker 2",
    "text": "This is like a template where you define a network allow list"
  },
  {
    "i": 111,
    "speaker": "Speaker 2",
    "text": "and any pre-install packages that you might want to have."
  },
  {
    "i": 112,
    "speaker": "Speaker 2",
    "text": "This is like the world that the agent is going to live in."
  },
  {
    "i": 113,
    "speaker": "Speaker 2",
    "text": "You take that, the environment and the agent,"
  },
  {
    "i": 114,
    "speaker": "Speaker 2",
    "text": "and you use that to run a session."
  },
  {
    "i": 115,
    "speaker": "Speaker 2",
    "text": "A sandbox is provisioned for you."
  },
  {
    "i": 116,
    "speaker": "Speaker 2",
    "text": "We set up the harness, and Cloud starts executing."
  },
  {
    "i": 117,
    "speaker": "Speaker 2",
    "text": "Any credentials and resources that you might want to include"
  },
  {
    "i": 118,
    "speaker": "Speaker 2",
    "text": "in that session are mounted into it"
  },
  {
    "i": 119,
    "speaker": "Speaker 2",
    "text": "and are available for the Cloud to use."
  },
  {
    "i": 120,
    "speaker": "Speaker 2",
    "text": "Finally, we have events."
  },
  {
    "i": 121,
    "speaker": "Speaker 2",
    "text": "These are anything that the agent produces"
  },
  {
    "i": 122,
    "speaker": "Speaker 2",
    "text": "as it's doing actions or any events that you might want"
  },
  {
    "i": 123,
    "speaker": "Speaker 2",
    "text": "to provide upstream in order to steer the agent."
  },
  {
    "i": 124,
    "speaker": "Speaker 2",
    "text": "This is how you might want to stay in the loop."
  },
  {
    "i": 125,
    "speaker": "Speaker 2",
    "text": "Events and stateful awareness is really where we're able"
  },
  {
    "i": 126,
    "speaker": "Speaker 2",
    "text": "to provide a platform so that you can build your own products"
  },
  {
    "i": 127,
    "speaker": "Speaker 2",
    "text": "on top of, and you can use primitives like memory"
  },
  {
    "i": 128,
    "speaker": "Speaker 2",
    "text": "and other features that we have in order"
  },
  {
    "i": 129,
    "speaker": "Speaker 2",
    "text": "to really optimize the performance of these agents."
  },
  {
    "i": 130,
    "speaker": "Speaker 1",
    "text": "So as Michael mentioned, events are the heart"
  },
  {
    "i": 131,
    "speaker": "Speaker 1",
    "text": "and the crux of what an agentic integration entails."
  },
  {
    "i": 132,
    "speaker": "Speaker 1",
    "text": "So let's demystify what that means in practice."
  },
  {
    "i": 133,
    "speaker": "Speaker 1",
    "text": "Everything in Managed Agents is event-based."
  },
  {
    "i": 134,
    "speaker": "Speaker 1",
    "text": "So these are durable transcripts that are clearly structured"
  },
  {
    "i": 135,
    "speaker": "Speaker 1",
    "text": "to help you track your agent's progress."
  },
  {
    "i": 136,
    "speaker": "Speaker 1",
    "text": "So first, there are user events."
  },
  {
    "i": 137,
    "speaker": "Speaker 1",
    "text": "These are actually what you're sending to the agent to guide it."
  },
  {
    "i": 138,
    "speaker": "Speaker 1",
    "text": "Next, there are agent events."
  },
  {
    "i": 139,
    "speaker": "Speaker 1",
    "text": "This is what the agent is actually doing."
  },
  {
    "i": 140,
    "speaker": "Speaker 1",
    "text": "So this is messaging."
  },
  {
    "i": 141,
    "speaker": "Speaker 1",
    "text": "This is tool execution, context compaction,"
  },
  {
    "i": 142,
    "speaker": "Speaker 1",
    "text": "engineering."
  },
  {
    "i": 143,
    "speaker": "Speaker 1",
    "text": "And even delegation to other agents."
  },
  {
    "i": 144,
    "speaker": "Speaker 1",
    "text": "Next, there are session events."
  },
  {
    "i": 145,
    "speaker": "Speaker 1",
    "text": "This is how you understand the progress of the unit of work"
  },
  {
    "i": 146,
    "speaker": "Speaker 1",
    "text": "that you've just delegated."
  },
  {
    "i": 147,
    "speaker": "Speaker 1",
    "text": "So this is the overall lifecycle, the status"
  },
  {
    "i": 148,
    "speaker": "Speaker 1",
    "text": "transitions, errors, and outcome process."
  },
  {
    "i": 149,
    "speaker": "Speaker 1",
    "text": "And lastly, there are span events."
  },
  {
    "i": 150,
    "speaker": "Speaker 1",
    "text": "There's a lot going on in this event stream."
  },
  {
    "i": 151,
    "speaker": "Speaker 1",
    "text": "This helps you group related events"
  },
  {
    "i": 152,
    "speaker": "Speaker 1",
    "text": "and just see things in a more aggregated, instrumented way."
  },
  {
    "i": 153,
    "speaker": "Speaker 2",
    "text": "So let's pivot to a real-world example of something"
  },
  {
    "i": 154,
    "speaker": "Speaker 2",
    "text": "that we built using"
  },
  {
    "i": 155,
    "speaker": "Speaker 2",
    "text": "Cloud Managed Agents, which is Pascal, which"
  },
  {
    "i": 156,
    "speaker": "Speaker 2",
    "text": "uses a hypothetical online grocery delivery services order"
  },
  {
    "i": 157,
    "speaker": "Speaker 2",
    "text": "data in order to analyze that data"
  },
  {
    "i": 158,
    "speaker": "Speaker 2",
    "text": "and provide insights for our team."
  },
  {
    "i": 159,
    "speaker": "Speaker 2",
    "text": "The agent produces analytics in minutes,"
  },
  {
    "i": 160,
    "speaker": "Speaker 2",
    "text": "leveraging a preloaded data set and a set of Python packages"
  },
  {
    "i": 161,
    "speaker": "Speaker 2",
    "text": "and scripts that we installed and uploaded into the container"
  },
  {
    "i": 162,
    "speaker": "Speaker 2",
    "text": "that it works from."
  },
  {
    "i": 163,
    "speaker": "Speaker 2",
    "text": "You can see every event in Cloud Console and even chat"
  },
  {
    "i": 164,
    "speaker": "Speaker 2",
    "text": "with a debug agent in order to further optimize"
  },
  {
    "i": 165,
    "speaker": "Speaker 2",
    "text": "your integration."
  },
  {
    "i": 166,
    "speaker": "Speaker 2",
    "text": "So going into the video itself, this"
  },
  {
    "i": 167,
    "speaker": "Speaker 2",
    "text": "is the Pascal kind of home page."
  },
  {
    "i": 168,
    "speaker": "Speaker 2",
    "text": "We can kick off an agent session and Cloud will get started."
  },
  {
    "i": 169,
    "speaker": "Speaker 2",
    "text": "It'll take it a couple of minutes to complete."
  },
  {
    "i": 170,
    "speaker": "Speaker 2",
    "text": "So while it does that, we'll go over to the developer console"
  },
  {
    "i": 171,
    "speaker": "Speaker 2",
    "text": "and we'll actually see the events that Cloud"
  },
  {
    "i": 172,
    "speaker": "Speaker 2",
    "text": "is producing in real time and be able to better understand"
  },
  {
    "i": 173,
    "speaker": "Speaker 2",
    "text": "what Cloud is doing live."
  },
  {
    "i": 174,
    "speaker": "Speaker 2",
    "text": "We can also actually look at the agent configuration"
  },
  {
    "i": 175,
    "speaker": "Speaker 2",
    "text": "and environment configuration that were set up for this session."
  },
  {
    "i": 176,
    "speaker": "Speaker 2",
    "text": "So this instance, that's the system prompt"
  },
  {
    "i": 177,
    "speaker": "Speaker 2",
    "text": "and that's the model."
  },
  {
    "i": 178,
    "speaker": "Speaker 2",
    "text": "And moving over to the environment configuration,"
  },
  {
    "i": 179,
    "speaker": "Speaker 2",
    "text": "you can see the network permissions that were allowed"
  },
  {
    "i": 180,
    "speaker": "Speaker 2",
    "text": "and the packages."
  },
  {
    "i": 181,
    "speaker": "Speaker 2",
    "text": "Jumping back to the actual page itself,"
  },
  {
    "i": 182,
    "speaker": "Speaker 2",
    "text": "it seems like our analysis is more or less done"
  },
  {
    "i": 183,
    "speaker": "Speaker 2",
    "text": "and we can actually dive into the insights that Cloud found."
  },
  {
    "i": 184,
    "speaker": "Speaker 2",
    "text": "On the product side, it seems like bananas are really,"
  },
  {
    "i": 185,
    "speaker": "Speaker 2",
    "text": "really popular, as well as a bunch of other products"
  },
  {
    "i": 186,
    "speaker": "Speaker 2",
    "text": "that are really popular, but everybody should get"
  },
  {
    "i": 187,
    "speaker": "Speaker 2",
    "text": "their bananas."
  },
  {
    "i": 188,
    "speaker": "Speaker 2",
    "text": "Looking into customer insights, we see that Sunday afternoons"
  },
  {
    "i": 189,
    "speaker": "Speaker 2",
    "text": "are really, really popular for online orders."
  },
  {
    "i": 190,
    "speaker": "Speaker 2",
    "text": "So you might want to schedule your orders"
  },
  {
    "i": 191,
    "speaker": "Speaker 2",
    "text": "for a different time of the day."
  },
  {
    "i": 192,
    "speaker": "Speaker 2",
    "text": "And then finally, we set up this prediction"
  },
  {
    "i": 193,
    "speaker": "Speaker 2",
    "text": "simulator that lets us analyze whether or not"
  },
  {
    "i": 194,
    "speaker": "Speaker 2",
    "text": "a customer is more likely to reorder things."
  },
  {
    "i": 195,
    "speaker": "Speaker 2",
    "text": "Jumping back to the developer console,"
  },
  {
    "i": 196,
    "speaker": "Speaker 2",
    "text": "we can see here in the side panel"
  },
  {
    "i": 197,
    "speaker": "Speaker 2",
    "text": "that we can kick off an analysis of the session itself, where"
  },
  {
    "i": 198,
    "speaker": "Speaker 2",
    "text": "Cloud will look at all of the events"
  },
  {
    "i": 199,
    "speaker": "Speaker 2",
    "text": "and further come up with insights for us"
  },
  {
    "i": 200,
    "speaker": "Speaker 2",
    "text": "on how to optimize the integration that we built."
  },
  {
    "i": 201,
    "speaker": "Speaker 2",
    "text": "In this particular instance, we see"
  },
  {
    "i": 202,
    "speaker": "Speaker 2",
    "text": "that some of the scripts that we gave Cloud are actually"
  },
  {
    "i": 203,
    "speaker": "Speaker 2",
    "text": "really, really slow."
  },
  {
    "i": 204,
    "speaker": "Speaker 2",
    "text": "So we may want to optimize the Python code that we wrote"
  },
  {
    "i": 205,
    "speaker": "Speaker 2",
    "text": "for them in order for them to do better."
  },
  {
    "i": 206,
    "speaker": "Speaker 2",
    "text": "So we just looked at a couple of things."
  },
  {
    "i": 207,
    "speaker": "Speaker 2",
    "text": "We looked at the developer console,"
  },
  {
    "i": 208,
    "speaker": "Speaker 2",
    "text": "and you saw that in action."
  },
  {
    "i": 209,
    "speaker": "Speaker 2",
    "text": "But there are really many different ways"
  },
  {
    "i": 210,
    "speaker": "Speaker 2",
    "text": "for you to get started with Cloud Managed Agents today"
  },
  {
    "i": 211,
    "speaker": "Speaker 2",
    "text": "in order to interact with these agents that you create."
  },
  {
    "i": 212,
    "speaker": "Speaker 2",
    "text": "You just pick whichever one fits your workflow best."
  },
  {
    "i": 213,
    "speaker": "Speaker 2",
    "text": "My favorite one is the Cloud API skill"
  },
  {
    "i": 214,
    "speaker": "Speaker 2",
    "text": "that we have available in Cloud Code today."
  },
  {
    "i": 215,
    "speaker": "Speaker 2",
    "text": "All you have to do is ask Cloud, hey,"
  },
  {
    "i": 216,
    "speaker": "Speaker 2",
    "text": "I want to get started with using Cloud Managed Agents,"
  },
  {
    "i": 217,
    "speaker": "Speaker 2",
    "text": "and it'll help you with the onboarding flow"
  },
  {
    "i": 218,
    "speaker": "Speaker 2",
    "text": "into your existing code bases."
  },
  {
    "i": 219,
    "speaker": "Speaker 2",
    "text": "Next, we also have the ANSI ally,"
  },
  {
    "i": 220,
    "speaker": "Speaker 2",
    "text": "which we released recently."
  },
  {
    "i": 221,
    "speaker": "Speaker 2",
    "text": "That makes it really, really easy to interact"
  },
  {
    "i": 222,
    "speaker": "Speaker 2",
    "text": "with our APIs for scripting and CI,"
  },
  {
    "i": 223,
    "speaker": "Speaker 2",
    "text": "CDL."
  },
  {
    "i": 224,
    "speaker": "Speaker 2",
    "text": "And then, lastly, we have our developer docks"
  },
  {
    "i": 225,
    "speaker": "Speaker 2",
    "text": "and our cookbooks."
  },
  {
    "i": 226,
    "speaker": "Speaker 2",
    "text": "These provide a lot of practical examples"
  },
  {
    "i": 227,
    "speaker": "Speaker 2",
    "text": "and copy-paste ready examples of the most common patterns"
  },
  {
    "i": 228,
    "speaker": "Speaker 2",
    "text": "that we have for Cloud Managed Agents."
  },
  {
    "i": 229,
    "speaker": "Speaker 1",
    "text": "So now that we've covered the basics,"
  },
  {
    "i": 230,
    "speaker": "Speaker 1",
    "text": "I want to touch on some of the more advanced features"
  },
  {
    "i": 231,
    "speaker": "Speaker 1",
    "text": "that we've recently shipped."
  },
  {
    "i": 232,
    "speaker": "Speaker 1",
    "text": "So first, multi-agent orchestration."
  },
  {
    "i": 233,
    "speaker": "Speaker 1",
    "text": "Here, Cloud is able to delegate tasks"
  },
  {
    "i": 234,
    "speaker": "Speaker 1",
    "text": "to other agents with independent context windows,"
  },
  {
    "i": 235,
    "speaker": "Speaker 1",
    "text": "enabling it to parallelize"
  },
  {
    "i": 236,
    "speaker": "Speaker 1",
    "text": "increasingly complex work."
  },
  {
    "i": 237,
    "speaker": "Speaker 1",
    "text": "With outcomes, Cloud iterates on predefined exit criteria"
  },
  {
    "i": 238,
    "speaker": "Speaker 1",
    "text": "or a rubric until it satisfies its goal."
  },
  {
    "i": 239,
    "speaker": "Speaker 1",
    "text": "You're in charge of this goal, and Cloud"
  },
  {
    "i": 240,
    "speaker": "Speaker 1",
    "text": "is in charge of completing it."
  },
  {
    "i": 241,
    "speaker": "Speaker 1",
    "text": "With memory, Cloud is able to read and write to memory stores."
  },
  {
    "i": 242,
    "speaker": "Speaker 1",
    "text": "By default, without memory, Cloud"
  },
  {
    "i": 243,
    "speaker": "Speaker 1",
    "text": "would be starting fresh on every session."
  },
  {
    "i": 244,
    "speaker": "Speaker 1",
    "text": "But with memory, it has an awareness of prior runs"
  },
  {
    "i": 245,
    "speaker": "Speaker 1",
    "text": "and can do better this next time."
  },
  {
    "i": 246,
    "speaker": "Speaker 1",
    "text": "Dreaming is built on top of memory,"
  },
  {
    "i": 247,
    "speaker": "Speaker 1",
    "text": "and here, Claude is reflecting and codifying"
  },
  {
    "i": 248,
    "speaker": "Speaker 1",
    "text": "on the learnings and codifying it into new memories."
  },
  {
    "i": 249,
    "speaker": "Speaker 1",
    "text": "And this ensures that it can continuously start"
  },
  {
    "i": 250,
    "speaker": "Speaker 1",
    "text": "with a tighter set of more curated and optimized memories."
  },
  {
    "i": 251,
    "speaker": "Speaker 1",
    "text": "All of these are really exciting intelligence features,"
  },
  {
    "i": 252,
    "speaker": "Speaker 1",
    "text": "and we're really excited to raise the ceiling"
  },
  {
    "i": 253,
    "speaker": "Speaker 1",
    "text": "of what agents can produce."
  },
  {
    "i": 254,
    "speaker": "Speaker 1",
    "text": "However, we also have heard that we need to meet you"
  },
  {
    "i": 255,
    "speaker": "Speaker 1",
    "text": "where you are, and that means making"
  },
  {
    "i": 256,
    "speaker": "Speaker 1",
    "text": "our infrastructure more modular."
  },
  {
    "i": 257,
    "speaker": "Speaker 1",
    "text": "So with self-hosted sandboxes, you can run the agent loop"
  },
  {
    "i": 258,
    "speaker": "Speaker 1",
    "text": "and have tool execution directly in your infrastructure"
  },
  {
    "i": 259,
    "speaker": "Speaker 1",
    "text": "so files and packages never leave your perimeter."
  },
  {
    "i": 260,
    "speaker": "Speaker 1",
    "text": "With MCP tunneling, Claude can access private MCP servers"
  },
  {
    "i": 261,
    "speaker": "Speaker 1",
    "text": "that you would not want to expose to the open internet."
  },
  {
    "i": 262,
    "speaker": "Speaker 1",
    "text": "Each of these things are built so that you can deploy"
  },
  {
    "i": 263,
    "speaker": "Speaker 1",
    "text": "within your enterprise, within your own security principles."
  },
  {
    "i": 264,
    "speaker": "Speaker 2",
    "text": "And in addition to all these amazing features"
  },
  {
    "i": 265,
    "speaker": "Speaker 2",
    "text": "that we already have, like Caitlin mentioned earlier today,"
  },
  {
    "i": 266,
    "speaker": "Speaker 2",
    "text": "we just released two new and very exciting features."
  },
  {
    "i": 267,
    "speaker": "Speaker 2",
    "text": "One is scheduled deployments, which allow you to set up"
  },
  {
    "i": 268,
    "speaker": "Speaker 2",
    "text": "a recurring schedule on which we'll trigger new sessions"
  },
  {
    "i": 269,
    "speaker": "Speaker 2",
    "text": "on your behalf for any recurring work that you might have."
  },
  {
    "i": 270,
    "speaker": "Speaker 2",
    "text": "And then next we have environment variables inside Vaults,"
  },
  {
    "i": 271,
    "speaker": "Speaker 2",
    "text": "which allow you to provide secure credentials"
  },
  {
    "i": 272,
    "speaker": "Speaker 2",
    "text": "for any APIs or CLIs that you might want Claude to call"
  },
  {
    "i": 273,
    "speaker": "Speaker 2",
    "text": "without actually having any sort of risk"
  },
  {
    "i": 274,
    "speaker": "Speaker 2",
    "text": "about Claude seeing the actual secret tokens."
  },
  {
    "i": 275,
    "speaker": "Speaker 2",
    "text": "Diving a little bit deeper into how environment variables work"
  },
  {
    "i": 276,
    "speaker": "Speaker 2",
    "text": "in Vaults, we place an opaque placeholder token"
  },
  {
    "i": 277,
    "speaker": "Speaker 2",
    "text": "inside the container itself that Claude has access to."
  },
  {
    "i": 278,
    "speaker": "Speaker 2",
    "text": "And whenever Claude is trying to hit an API or use a CLI,"
  },
  {
    "i": 279,
    "speaker": "Speaker 2",
    "text": "it will just use that environment variable the way"
  },
  {
    "i": 280,
    "speaker": "Speaker 2",
    "text": "that it would use any other environment variable."
  },
  {
    "i": 281,
    "speaker": "Speaker 2",
    "text": "And as that network request is made,"
  },
  {
    "i": 282,
    "speaker": "Speaker 2",
    "text": "we will inject the real secret value as the request is made"
  },
  {
    "i": 283,
    "speaker": "Speaker 2",
    "text": "so that Claude may never actually see these."
  },
  {
    "i": 284,
    "speaker": "Speaker 2",
    "text": "So that's really the value of the secret token."
  }
]