Claude Cowork comes to Amazon Bedrock for org-wide agents

TLDR: Claude Cowork lands on Amazon Bedrock, AWS ships an ML lineage recipe, and Havas lays out a very opinionated agentic marketing stack.

Claude Cowork and Claude Code Desktop now run on Amazon Bedrock

Amazon Web Services and Anthropic are bringing Claude Cowork and Claude Code Desktop into Amazon Bedrock, so you can run these agentic tools directly against your existing Bedrock setup as of 2026-04-22. The integration lets teams access Claude Cowork through Bedrock or an existing large language model (LLM) gateway, instead of managing separate vendor-specific tooling.

For AI platform teams already standardizing on Amazon Bedrock, this reduces friction: authentication, logging, observability, and guardrails can be centralized while knowledge workers and developers use Claude-based agents for day to day tasks. The blog hints at workflows where Claude Cowork runs against internal data, but the security and governance story will depend on how you wire Bedrock into your own data plane.

If your org is already invested in Amazon Web Services and evaluating agentic assistants for enterprise rollout, this is a concrete way to pilot Claude at team and org scale without adding another SaaS silo.

Read more →

---

AWS details end to end ML lineage with DVC and SageMaker AI

Amazon Web Services published a reference pattern that combines Data Version Control (DVC), Amazon SageMaker AI, and Amazon SageMaker AI MLflow Apps to implement end to end machine learning model lineage as of 2026-04-22. The post walks through two concrete setups: dataset level lineage and record level lineage, with companion notebooks you can run in your own account.

This matters if you are trying to operationalize responsible AI or just keep auditors and compliance teams off your back. DVC tracks data snapshots, while SageMaker AI and MLflow Apps capture experiment metadata and model deployment history so you can trace which data produced which model and which model served which prediction. Record level lineage is more complex and will not be cheap at scale, but it gives far better answers when something goes wrong in production.

Teams already half using MLflow and half using Amazon SageMaker often end up with fragmented metadata; this pattern is a useful starting point, although you will still need to adapt it to your own feature store and CI setup.

Read more →

---

Havas reveals agentic marketing stack built on Converged.AI

Havas is outlining its Havas Converged.AI ecosystem, a suite of agentic products positioned as a "human led agentic ecosystem" for marketing and communications as of 2026-04-22. The stack is organized into four phases: Intelligence, Design, Activate, and Measure, and uses intelligent agents that respond to natural language and execute code to replace high friction manual workflows.

For anyone building agents for non technical business users, this is a useful case study in how a large agency packages and narrates an agentic product suite instead of a collection of point tools. The architecture leans on partners and data orchestration rather than a single monolithic platform, which is closer to how most enterprises will actually deploy. The language is still marketing heavy and does not include latency, reliability, or safety metrics, so treat it as a directional blueprint rather than a technical spec.

If you are selling or designing agent platforms for marketing, Havas is effectively showing the reference buyer story your clients will start to expect.

Read more →

---

Quick Hits

• Intelligence, Rearranged: How Agents Are Changing Legal Work Harvey is reported to have run over 700,000 agentic tasks in legal workflows, shifting from rigid checklists to more flexible decision making. Good perspective if you are targeting professional services where risk and explainability dominate the buying criteria.

• Agentic Hospitality: AI-Driven Discovery + TravelOS Agentic Hospitality outlines a travel stack built on Google Cloud Vertex AI, TravelOS, and marketing tools, arguing that new tracking models from Google require metrics beyond traditional hotel KPIs.

• Show HN: Graph Compose – Temporal workflows with visual builder, SDK, and AI Graph Compose orchestrates API workflows on Temporal with a React Flow visual builder, a typed TypeScript SDK, and an AI assistant that turns natural language into graphs. Open core with AGPL foundations, useful if your agents need reliable long running workflows instead of ad hoc cron jobs.

• Show HN: GoModel – an open-source AI gateway in Go GoModel is an open source AI gateway in Go that sits between your app and providers like OpenAI and Anthropic, adding usage tracking, cost allocation, semantic caching, and easy model switching. The lightweight Docker image and focus on observability make it interesting if you want a self hosted alternative to commercial gateways.

• Scaling Codex to enterprises worldwide OpenAI launches Codex Labs and partnerships with Accenture, PwC, Infosys and others to roll out Codex tooling across enterprise software development, reporting 4 million weekly active users.

• AI and the Future of Cybersecurity: Why Openness Matters Hugging Face discusses the implications of Mythos and Project Glasswing for global cybersecurity and makes the case for open tooling and transparency instead of purely closed frontier systems.

• GPT Image 2 on AI Gateway Vercel AI Gateway now exposes GPT Image 2 with strong instruction following, fine grained UI element rendering, coherent non English text, and resolutions up to 2K, which is handy for any agent that needs to generate production ready visuals.

• Show HN: Daemons – we pivoted from building agents to cleaning up after them Charlie Labs describes a pivot from a TypeScript coding agent to tools that monitor and clean up agent side effects, reflecting a broader shift from "build the agent" to "govern and contain the agent".

• Partnering with industry leaders to accelerate AI transformation Google DeepMind announces partnerships with global consultancies to bring frontier AI into enterprises, reinforcing that the consulting channel is becoming the default route for large scale AI transformation projects.