CHAPTERS
- 0:13 – 0:43
Session goal: moving Claude Code prototypes to production on AWS
Antonio Rodriguez (AWS) frames the session as a hands-on path from local Claude Code prototypes to production-ready cloud deployments. He sets expectations: a short partnership overview followed by guided workshop time.
- •Positioning: developers now “orchestrate” models more than write code
- •Focus on productionizing Claude Code-built apps on AWS
- •Hands-on format with guided instructions and workshop environment
- •Brief intro first, then practical build time
- 0:43 – 2:45
AWS + Anthropic partnership and long-term investment context
The talk highlights the multi-year collaboration between AWS and Anthropic, emphasizing a “better together” narrative. It cites mutual commitments and why the partnership matters for customers building with Claude at scale.
- •Relationship timeline from early Claude generations to today
- •Multi-billion Amazon investment in Anthropic
- •Anthropic’s large-scale commitment to AWS infrastructure usage
- •Theme: combining frontier models with cloud platform capabilities
- 2:45 – 3:15
Infrastructure behind Claude on AWS: Project Rainier and Trainium
Antonio points to the underlying compute that supports training and hosting Claude models, focusing on dedicated infrastructure and AWS’s purpose-built chips. The emphasis is performance and cost efficiency for token usage.
- •Project Rainier as major AI compute infrastructure for Claude
- •Purpose-built AWS Trainium chips for Claude workloads
- •Third-generation Trainium mentioned for speed/cost benefits
- •Framing: infrastructure choices improve inference economics
- 3:15 – 6:17
Three technical pillars: platform features, security/privacy, and scalability
The session shifts to the core value propositions of using Claude via AWS: a rich platform layer (Bedrock features), strong security/privacy controls, and flexible scaling and regional deployment. These pillars motivate why teams choose AWS for production deployments.
- •Feature breadth around the model (tooling + integrations)
- •Security posture: keep data within AWS boundaries
- •Privacy: avoid public internet “data hopping”
- •Scalability and region choice for regulatory needs (e.g., GDPR)
- 6:17 – 7:18
Bedrock capabilities beyond models: evaluation, prompt optimization, RAG, and guardrails
Antonio surveys Amazon Bedrock’s broader developer toolkit: evaluation and prompt optimization workflows, fine-tuning/distillation, managed knowledge bases for RAG, and guardrails to reduce risk. The goal is end-to-end support for building reliable applications.
- •Model evaluation and prompt optimization features
- •Fine-tuning and distillation options (including mention of Haiku fine-tuning)
- •Managed knowledge bases for RAG-style applications
- •Guardrails for topic filtering, PII masking, and hallucination controls via grounding/checks
- 7:18 – 7:49
Agentic AI on AWS: Bedrock Agent Core and framework compatibility
The talk connects Claude’s agent workflows to AWS hosting and orchestration options. Bedrock Agent Core is presented as an infrastructure layer that works with popular agent frameworks and the Claude Agent SDK.
- •Rise of agentic AI as a major focus area
- •Bedrock Agent Core for hosting agents securely in the cloud
- •Compatibility with common frameworks (e.g., LangChain)
- •Support for Claude Agent SDK-based implementations
- 7:49 – 9:50
“Why Claude on AWS” recap: sovereignty, billing, observability, SLAs, identity, private networking
Antonio summarizes the “better together” benefits in practical enterprise terms. The emphasis is operational readiness: governance, cost management, monitoring, reliability, authentication, and private connectivity.
- •Data sovereignty and controlled deployment environment
- •Centralized/consolidated AWS billing for Claude usage
- •Observability via CloudWatch/CloudTrail (metrics, logs, traces)
- •Production assurances with SLAs and availability expectations
- •Identity integration (SSO/OAuth/IAM; Entra ID/Okta examples) and PrivateLink to avoid public internet exposure
- 9:50 – 10:21
Compliance and regulated-industry readiness (FedRAMP, HIPAA, etc.)
The session briefly highlights compliance programs supported in the AWS environment, positioning Claude on Bedrock for regulated sectors. Attendees are directed to online documentation for the full list and details.
- •Compliance alignment for regulated workloads
- •Examples cited: FedRAMP and HIPAA
- •Reinforces security/privacy claims from earlier sections
- •Encouragement to consult AWS compliance documentation
- 10:21 – 10:51
Three ways to use Claude on AWS (1): Directly via Amazon Bedrock APIs
Antonio lays out the first access path: invoking Claude models through Bedrock, including updated model availability and API options. This path emphasizes AWS-native controls and integrations while maintaining a Claude experience.
- •Claude available through Bedrock as a primary integration route
- •Mentions model availability updates (e.g., Opus 4.7)
- •Messages API support for direct invocation style preferences
- •AWS-native governance and ecosystem integration remain in place
- 10:51 – 11:51
Three ways to use Claude on AWS (2): Claude platform on AWS (feature-parity gateway)
The second option is the Claude platform delivered through AWS, described as a gateway that retains Anthropic feature parity while using AWS billing and access control. It’s positioned as “best of both worlds” for organizations standardizing on AWS governance.
- •Claude platform on AWS described as newly GA
- •AWS provides consolidated billing and access control
- •Inference requests are routed to Anthropic for processing
- •Feature parity called out: web search, files, agents, etc.
- 11:51 – 12:51
Three ways to use Claude on AWS (3): Claude Desktop/Claude Code with AWS billing and connectors
The third path targets end-user tooling: Claude Desktop and Claude Code workflows that can be paid through AWS and connected to Bedrock or the Claude platform on AWS. Options include AWS Marketplace (enterprise seats) and direct connection via Anthropic’s connector capabilities.
- •Support for Claude Desktop and Claude Code usage with AWS billing
- •Claude Enterprise via AWS Marketplace for seat-based subscriptions
- •Option to connect Claude Desktop to Bedrock or Claude platform on AWS
- •Goal: meet teams where they already work (desktop + terminal workflows)
- 12:51 – 14:53
Workshop overview: prerequisites, modules, and what you’ll build (Calidraw + automation)
Antonio transitions to the hands-on workshop, listing prerequisites (AWS account provided; Claude Code installed) and describing the workshop project using the Calidraw repository. He outlines module progression from basics to advanced automation, git workflows, and agent-like extensibility.
- •Prereqs: workshop AWS account environment + Claude Code installed
- •Workshop uses Calidraw repo to generate/modify architecture diagrams
- •Module 1: setup and basic Claude Code usage
- •Module 2: context work + Playwright MCP for screenshots/automated edits + git workflows
- •Module 3: sub-agents/plugins + custom skills/hooks + advanced configuration
- 14:53 – 15:54
Advanced/production workshop track: team standards, cost control, dashboards, and Agent SDK on AWS
He points attendees to a more advanced workshop designed for teams moving to production. Topics include governance and standardization, scaling and cost controls, usage dashboards, measuring productivity/ROI, and deploying agent solutions using Bedrock Agent Core.
- •Team standards and distributing advanced workflows
- •Scalability practices and cost-control mechanisms
- •Dashboards for token usage and operational visibility
- •Developer productivity and ROI measurement ideas
- •Using Claude Agent SDK with Bedrock Agent Core for production agent deployments
- 15:54 – 17:56
Getting access: workshop accounts, safety rules, and environment options (hosted VS Code or your machine)
Antonio explains how to join the workshop event, accept terms, and access a prebuilt environment (terminal + VS Code UI). He also describes alternative setups: running Claude in a local terminal or using Claude Desktop connected to Bedrock, plus references to a supporting blog.
- •Join flow includes terms: don’t upload confidential/PII data
- •Provisioned environment includes URL to terminal + VS Code UI
- •Alternative: run locally and configure Claude Code to point to Bedrock
- •Option: Claude Desktop connected to Bedrock for the same workshop flow
- •Reference guidance/blog for Claude Code + desktop tooling on Bedrock
- 17:56 – 19:41
Step-by-step join instructions and support, plus key Claude Code configuration knobs
The session closes with practical access steps (join.workshop.aws + code) and in-room support instructions. Antonio highlights advanced configuration parameters for Claude Code—region, Bedrock endpoint/model defaults, token/rate controls, telemetry, and editor extensions to reduce friction.
- •Access method: join.workshop.aws and enter the event code
- •In-room AWS solutions architects available for help
- •Configuration: set region, Bedrock target, and default model
- •Controls: token usage reduction, rate limiting, telemetry auto-reporting
- •Developer experience tweaks via extensions (disable login prompt/onboarding messages)
