PLUS: Supersonic turbines for AI data centers, Mistral's Devstral 2, and AI discovering algorithms
Good morning
In a significant and unexpected move, major AI rivals OpenAI and Anthropic are joining forces. The two companies have teamed up with Block to launch a new foundation aimed at building an open, collaborative ecosystem for AI agents to work together seamlessly.
The initiative aims to prevent a fragmented landscape by creating common ground for the entire industry. But can this alliance truly overcome the intense competitive dynamics of the AI race to create a unified and interoperable future for autonomous agents?
In today’s Next in AI:
OpenAI and Anthropic unite for agentic AI
Supersonic turbines to power AI data centers
Mistral’s new open-source coding model
An AI framework that discovers new algorithms
AI's Titans Unite
Next in AI: In a landmark move, fierce rivals OpenAI, Anthropic, and Block have joined forces to launch the Agentic AI Foundation under the Linux Foundation. Their goal is to build an open, collaborative ecosystem for AI agents, ensuring they can work together seamlessly.
Decoded:
This initiative brings together the industry's heaviest hitters, with the three founding members backed by platinum members including Google, Microsoft, AWS, and Bloomberg, signaling strong industry-wide support.
The foundation launches with three key projects: Block's agent framework goose, OpenAI's coding agent guide AGENTS.md, and Anthropic's Model Context Protocol (MCP) for connecting AI to external systems.
To show the immediate impact, MCP alone already powers over 10,000 active servers and sees more than 97 million monthly SDK downloads, demonstrating the scale these open standards will operate on.
Why It Matters: This collaboration aims to prevent a fragmented AI landscape by creating common ground, much like open standards did for the internet. For developers and businesses, this signals a major push toward a future where autonomous agents from different companies can easily communicate and cooperate.
Supersonic AI Power
Next in AI: Boom Supersonic is repurposing its jet engine technology to create 'Superpower,' a new turbine designed to solve the energy crisis facing AI data centers. This initiative directly addresses the electricity shortage currently slowing AI's expansion.
Decoded:
Current data center turbines are based on outdated 1970s jet engines that lose up to 30% of their power in hot weather, a major issue for popular data center locations like Texas.
Unlike legacy systems, Superpower's design is based on a supersonic engine core, allowing it to maintain full 42MW output even at 110°F without needing water for cooling.
The company is fast-tracking production by vertically integrating its supply chain and already has a 1.21-gigawatt order from launch customer Crusoe.
Why It Matters: Boom is cleverly turning its engine technology into a new revenue stream that solves a critical bottleneck for the entire AI industry. This move provides a self-sustaining path to funding its supersonic airliners while validating its core engine tech in the process.
Mistral's Open-Source Coder
Next in AI: Mistral AI released Devstral 2, a new family of high-performance, open-source coding models designed to run everywhere from data centers to consumer hardware. The models are currently free to use via Mistral's API and come with a new command-line agent for automating development tasks.
Decoded:
The larger 123B model sets a new open-source record with 72.2% on SWE-bench Verified, while being significantly smaller and more cost-efficient than its competitors.
A smaller 24B version brings powerful coding capabilities to consumer hardware, enabling fast, private, on-device runtimes. This is paired with Mistral Vibe CLI, a new command-line agent you can install as a Zed IDE extension to automate development.
The model is built for production workflows, able to orchestrate changes across complex codebases and fix bugs. It can be fine-tuned for specific languages or enterprise systems and is available for deployment on NVIDIA's build platform.
Why It Matters: This release significantly narrows the performance gap between open-source and proprietary coding models. Developers now have more powerful and accessible tools to build and automate, right from their local machines.
AI Discovers Algorithms
Next in AI: ASI Labs has released OpenEvolve, an open-source framework that uses large language models to automatically discover new, high-performance algorithms. The system applies an evolutionary process to generate, test, and refine code, automating a task that typically requires deep human expertise.
Decoded:
OpenEvolve works by creating a diverse population of potential solutions and iteratively improving them. It uses a quality-diversity search to prevent settling on a single good idea, instead exploring many different and effective algorithmic paths simultaneously, much like natural selection.
The framework has already shown impressive results in GPU kernel optimization, discovering novel techniques to improve transformer performance on Apple Silicon hardware. It automatically found non-obvious solutions, such as 8-element SIMD vectorization, that led to measurable speedups while maintaining 100% numerical accuracy.
It's not just for traditional code; the system can also evolve prompts themselves to improve LLM reasoning. On benchmarks, its evolved prompts increased accuracy by over 10% on multi-hop reasoning tasks, demonstrating the framework's broad utility.
Why It Matters: Tools like this are shifting the developer's role from writing code to curating and guiding an AI-driven discovery process. This approach has the potential to unlock novel solutions for complex problems in science and engineering.
AI Pulse
Accenture expanded its partnership with Anthropic, forming a dedicated business group to train 30,000 professionals on Claude and create one of the world's largest ecosystems of Claude practitioners for enterprise AI deployment.
The EU launched an investigation into Google over concerns its AI Overviews use content from publishers and YouTube creators without appropriate compensation or the ability for them to opt-out.
AI emerged as the frontrunner for Time's 2025 Person of the Year, with prediction markets giving it over a 40% chance of winning, ahead of figures like Nvidia CEO Jensen Huang and OpenAI CEO Sam Altman.