🧭 Claude Platform Comes to AWS: Full Managed Agents, Code Execution & MCP Through Your AWS Account
Anthropic has launched Claude Platform on AWS, making its full native platform available directly inside AWS accounts via IAM permissions and AWS billing — no separate Anthropic account or API key required. This is distinct from Claude models on Amazon Bedrock: Bedrock provides model access; Claude Platform on AWS delivers the entire agentic layer — managed agents, code execution, web search, file handling, and Model Context Protocol (MCP) connectors — as a managed AWS service.
What "Claude Platform on AWS" actually includes
- Managed agent infrastructure — Claude runs multi-step agentic tasks (web search, code execution, file I/O, tool calls) within AWS's security boundary. Agents are defined, invoked, and monitored through AWS APIs with full CloudTrail audit logging.
- MCP connector registry — Pre-built MCP connectors for popular enterprise systems (Jira, Confluence, Salesforce, GitHub, S3, RDS) deploy in minutes. Custom connectors are deployed as Lambda functions within your VPC.
- Code execution sandbox — Secure Python/JavaScript execution environment that integrates with your existing AWS compute (EC2, ECS, Lambda). Output goes to S3; no data leaves your AWS region.
- Unified AWS billing — Claude usage rolls up into your existing AWS invoice. Organizations already using AWS Enterprise Discount Program credits apply them to Claude usage automatically.
Why this matters for enterprise procurement
The single biggest friction point in enterprise AI adoption is not the technology — it's procurement, security review, and billing consolidation. By routing Claude through AWS, teams that already have approved AWS vendor relationships and compliance frameworks can deploy Claude without a separate security review cycle. If your org is AWS-native and you've been waiting for Claude to clear InfoSec, this removes that blocker. Start with a test project in a dev AWS account before requesting production access.
# Invoking a Claude managed agent via AWS SDK (Python)
import boto3
claude = boto3.client("claude-platform", region_name="us-east-1")
response = claude.invoke_agent(
agentId="my-research-agent",
sessionId="session-abc123",
inputText="Summarise last quarter's support tickets from Zendesk",
enableTrace=True
)
for event in response["completion"]:
if "chunk" in event:
print(event["chunk"]["bytes"].decode(), end="", flush=True)
AWS
Claude Platform
managed agents
MCP
enterprise
cloud native
IAM
🧭 Anthropic in Advanced Talks to Acquire Stainless — the SDK Factory Behind OpenAI, Google & Meta's Libraries
Anthropic is in advanced acquisition talks for Stainless, a New York-based developer tools startup that automatically generates production-ready SDKs from OpenAPI specifications, according to The Information. The deal is valued at over $300 million. If it closes, Anthropic will own the toolchain that currently generates and maintains the official client libraries for OpenAI, Google, and Meta — including the widely used pip install openai Python package.
What Stainless does — and why it matters
Stainless ingests an OpenAPI spec and produces idiomatic, well-tested SDKs in Python, TypeScript, Go, Java, Ruby, and more — with pagination helpers, retry logic, streaming support, and full type annotations — automatically. For AI API providers, maintaining SDKs across six languages is a significant engineering burden; Stainless turns it into a configuration file.
- Strategic infrastructure: Owning Stainless gives Anthropic control over a layer that every AI SDK consumer touches daily, regardless of which model they're calling.
- SDK quality advantage — Anthropic's own SDKs (
anthropic-sdk-python, @anthropic-ai/sdk) are already built on Stainless. Bringing it in-house would accelerate Claude Code's ability to generate correct client code for any API.
- Competitive sensitivity — Stainless also generates OpenAI's and Google's SDKs. Post-acquisition, the competitive governance of that relationship would be complex; Anthropic would likely continue serving those contracts while investing most engineering focus on Claude-adjacent tooling.
This is still in talks — not closed
The Information's reporting notes "advanced talks," not a signed deal. The acquisition could fall through, be renegotiated, or face regulatory review. Treat this as a strategic signal about Anthropic's direction — not a confirmed product change. If you depend on Stainless-generated SDKs in your stack (for OpenAI or otherwise), there's nothing to change today. Watch for an official announcement from either company.
Stainless
acquisition
SDK tooling
developer tools
OpenAPI
M&A
🧭 Salesforce's Marc Benioff: $300M Claude Token Spend in 2026, Driven by Coding Agents
Salesforce CEO Marc Benioff disclosed on the All-In podcast that Salesforce plans to spend approximately $300 million on Anthropic Claude tokens in 2026, making it one of the largest known single-company Claude deployments. The spend is concentrated on coding agents — Benioff described Claude Code and Claude-powered coding workflows as transforming how Salesforce's engineering teams build and ship product. He also revealed that Microsoft blocked Salesforce from investing in OpenAI, which drove Salesforce's decision to take a $330 million equity stake in Anthropic (roughly 1% at Anthropic's then-$380 billion valuation).
What $300M in tokens reveals about enterprise AI economics
- Coding is the anchor use case — Benioff was explicit that the spend is disproportionately on coding rather than customer-facing AI features. This validates the pattern seen in PwC's deployment: the ROI on AI-assisted software development is outrunning other enterprise AI applications.
- Token cost and productivity — At Claude's API pricing (Sonnet 4.6 input/output roughly $3/$15 per million tokens), $300M represents an enormous volume of coding context flowing through Claude. Benioff implied that the savings in engineering time and headcount far exceed the token cost.
- Strategic equity + commercial spend — Salesforce holds both equity in Anthropic and a large commercial commitment. This "investor-customer" structure is becoming a pattern: enterprises with enough AI spend to negotiate equity stakes are doing so, locking in pricing and alignment with the model provider's roadmap.
The "so what?" for developers
If Salesforce — a company with mature internal engineering tooling and thousands of engineers — is committing $300M to Claude coding agents, the economics of AI-assisted development have crossed a threshold that's hard to ignore. The calculation is: developer productivity gain × engineering headcount × average loaded salary, compared against token cost. At scale, the ratio appears to be favourable enough that even token-cost-conscious CFOs are approving budgets of this size. For teams still in "pilot mode" with Claude Code, Benioff's disclosure is useful evidence for internal budget conversations.
Salesforce
Marc Benioff
enterprise spend
coding agents
Claude Code
token economics
investor-customer