Global AI Weekly
Issue number: 147 | Tuesday, April 28, 2026
Highlights
Anthropic and AWS: A $100B Bet on Compute as Strategy
Amazon’s $5 billion investment in Anthropic comes with a striking counterweight: a commitment to spend more than $100 billion on AWS over the coming decade. The deal reflects a deeper shift in AI, where access to large-scale compute is becoming the defining constraint. Capital is no longer just funding innovation; it is securing infrastructure, binding model labs and cloud providers into long-term, mutually dependent relationships.
anthropic.com
OpenAI turns on cost-per-click ads inside ChatGPT
OpenAI has started rolling out cost-per-click (CPC) ads within ChatGPT as part of its efforts to explore advertising opportunities on the platform. This shift aligns with the company's plans to bring on board its first Advertising Marketing Science Lead, signaling a deeper focus on ad-based revenue models. This marks a significant step in how OpenAI monetizes its AI-driven chatbot.
digiday.com
Changes to GitHub Copilot plans for individuals
GitHub’s latest changes to Copilot plans feel like a quiet turning point. New signups for paid individual tiers are paused, usage limits are tighter, and the most powerful models are now gated behind higher pricing tiers. The reason is simple and telling: AI coding has outgrown flat-rate pricing. Agentic workflows, long-running sessions, and parallel tasks are burning far more compute than Copilot was ever designed for. What this really signals is a shift from “AI as a feature” to AI as metered infrastructure. If you’re building with these tools, expect more limits, more pricing tiers, and fewer illusions about “unlimited” AI.
github.blogResearch
OpenGame: Open Agentic Coding for Games
This paper page introduces OpenGame, a framework designed for open agentic coding in the realm of gaming. It explores innovative approaches and methodologies to enhance the development and interaction of agents within games. The discussion aims to broaden understanding and foster collaboration in this area of technology.
huggingface.co
Reward Hacking in the Era of Large Models: Mechanisms, Emergent Misalignment, Challenges
This paper explores the concept of reward hacking in the context of large machine learning models, examining how these models can exploit or misalign with intended objectives. It highlights the underlying mechanisms driving this behavior, the challenges it poses for alignment, and potential implications for future AI systems. The discussion aims to shed light on emergent issues tied to the growing complexity of modern AI models.
huggingface.coVideo
Building your own MCP server
In Episode 5 of the Made for Dev Show, host Simon teams up with Microsoft MVP and Software Engineer Stuart to explore building a Model Context Protocol (MCP) server using the C# SDK. The episode breaks down MCP's core components—tools, resources, and prompts—while offering a hands-on demonstration of a Dunder Mifflin-themed server. Stuart showcases local and remote server setups, integration with tools like GitHub Copilot, and real-world use cases, including retrieving database details and attaching resources like employee handbooks. It’s a practical guide for developers interested in creating AI-driven solutions with MCP.
youtube.com
Build Your Own Deep Research Agents
This workshop offers a hands-on experience in creating a deep research agent powered by MCP, capable of planning, analyzing content, gathering evidence, and synthesizing findings into a structured research artifact. Attendees will explore the integration of this agent into a streamlined technical writing workflow, learning how research and writing demand distinct systems for maximum efficiency. The session focuses on practical system design, balancing exploratory AI capabilities with refined, guided workflows for high-quality output. Participants will walk away with a functional research agent, insights into engineering trade-offs, and tools for measurable and improvable AI systems.
youtube.comArticles
Ranked: AI Models U.S. Businesses Pay For
OpenAI remains a top choice for U.S. businesses investing in AI, but Anthropic is catching up with innovative offerings like Claude Code and Cowork. These tools are gaining traction, showcasing a growing competition in the paid AI services space. As more companies look to integrate AI, the market is shaping up with diverse options tailored to different needs.
visualcapitalist.com
Building AI agents without the cloud
The idea that you need massive models and endless API credits to build useful AI is quietly breaking down. This hands-on guide shows how to build a fully functional AI agent locally using small language models, running entirely on your own machine. Using tools like Ollama and LangChain, it walks through how to add memory, tools, and decision-making capabilities, turning a simple model into something that can actually act, not just respond. The bigger shift is what this represents: AI is becoming personal infrastructure. Cheaper, private, and increasingly good enough. Not as powerful as frontier models, but far more practical for real-world use.
machinelearningmastery.com
Secure-by-design: 3 principles to safely scale agentic AI
Autonomous AI agents are revolutionizing enterprises with their efficiency and capabilities, but they also come with significant security challenges. Ensuring these AI systems are secure-by-design is crucial to mitigate risks that can escalate rapidly. This approach involves adhering to key principles that prioritize safety while enabling organizations to scale their use of agentic AI effectively.
cio.comUpcoming Events
WeAreDevelopers World Congress 2026
The WeAreDevelopers World Congress 2026 takes place July 8–10 in Berlin, bringing together 15,000+ developers and AI leaders. This is less about hype and more about how AI is actually used in production. Expect focused sessions on LLMs, agentic systems, and AI-driven development, alongside hands-on workshops and strong technical depth. A solid pick if you’re building real-world AI systems, not just experimenting.
AgentCamp - Coming to a City Near You
AgentCamp continues to grow as a global series of hands-on gatherings dedicated to building and experimenting with AI agents. These community-driven events bring developers, founders, and AI enthusiasts together for practical sessions, collaborative building, and open exchange of ideas. Hosted in cities around the world, AgentCamp focuses on real-world experimentation, giving participants the space to prototype agent workflows, explore emerging tools, and learn directly from peers working at the edge of autonomous AI. Join the community to build, share, and help advance what AI agents can do in practice.
globalai.communityCode
GitHub - google-labs-code/design.md: A format specification for describing a visual identity to coding agents
Google Labs' DESIGN.md is a format specification aimed at providing coding agents with a clear and structured way to understand and implement a design system. It standardizes how visual identities are described, ensuring consistency and persistence across projects. This tool is particularly useful for bridging the gap between designers and developers, streamlining the collaboration process.
github.com
modelcontextprotocol/mcpb: Desktop Extensions: One-click local MCP server installation in desktop apps
Discover the convenience of Desktop Extensions with a one-click solution for installing local MCP servers directly within desktop applications. This tool simplifies setup, making it easier to integrate and manage local server environments effortlessly. Ideal for developers seeking a streamlined approach to working with MCP servers on their desktops.
github.com
Claude Code + Cloudflare
A terminal-based coding agent designed to assist with codebase navigation, running commands, editing files, and handling Git tasks. Created by Anthropic, it streamlines development workflows by integrating seamlessly into your coding environment.
developers.cloudflare.comPodcast
How I AI
"How I AI," hosted by Claire Vo, offers a practical and approachable guide to using AI tools to boost productivity and enhance work quality. Each 30-minute episode features guests sharing real-world examples, workflows, and actionable tips that listeners can apply right away. With live screen sharing and easy-to-follow advice, the podcast is perfect for anyone looking to build confidence and skills in navigating the ever-evolving world of AI.
open.spotify.com