Good morning. It’s Friday, May 22nd.

You read. We listen. Let us know what you think by replying to this email.

^{In partnership with Intology}

Most coding agents are great at cleanup work. They patch bugs. Refactor code. Tune parameters. But when the task requires real algorithmic insight, they usually stall, overthink, or make things worse.

Intology is trying to change that.

Their Artificial Scientist, Locus, recently set a new world record on NanoGPT-Bench, a benchmark designed to separate real AI R&D from low-hanging fruit. The breakthrough came from implementing a fused Triton kernel for a difficult training step, pushing past a benchmark where humans had already set a very high bar.

That matters because the future of AI research won’t be won by agents that just polish existing code. It will be won by systems that can push the frontier.

Intology is hiring builders who want to work on that frontier.

Learn more and apply here

Alibaba launched Qwen3.7-Max, a proprietary flagship model capable of 35 hours of continuous autonomous execution. Trained via environment scaling, it handles long-horizon reasoning without logic loops, executing 1,158 tool calls to achieve a 10.0x kernel optimization speedup on unfamiliar hardware.

The model offers a 1-million-token context window, supports cross-harness generalization to plug natively into external tools like Claude Code, and beats Claude Opus-4.6 Max on key math benchmarks. Pivoting away from its open-source roots, Alibaba restricts this release to a paid API tier, positioning it as a more affordable enterprise alternative to Western frontier models. Read more.

Google has tripled weekly quotas across all paid Antigravity plans after a sharp rise in developer usage overwhelmed existing limits. The move is meant to prevent interruptions during long AI-assisted coding sessions and support sustained, high-volume development workflows.

For enterprise-scale agents, Google also released an open-source runtime called Agent Executor (AX). Built on Kubernetes, AX uses kernel snapshots and state recovery to pause, recover, and stabilize broken multi-agent loops, forming the infrastructure behind advanced systems like the Nexus forecasting engine.

Google is also expanding agentic automation into business and creative tools. Pomelli can now analyze files and images to build a “Business DNA” profile, then automatically generate full brand books and launch complete websites. Meanwhile, Gemini now connects directly with Adobe, Canva, and CapCut, allowing users to edit layered designs, images, and videos through conversational prompts inside a single workspace.

At the web infrastructure layer, Google introduced an “Agentic Browsing” audit in Lighthouse. The tool checks whether websites are machine-readable and stable enough for autonomous AI navigation. Read more.

Anthropic is breaking the industry’s cash-burn narrative, projecting a historic $10.9 billion Q2 2026 revenue, a 130% surge over Q1, and its first operating profit of $559 million. This massive scalability, fueled by deep enterprise adoption across Fortune 10 firms, has dropped compute overhead to 56 cents per revenue dollar, pushing private valuation talks toward $900 billion.

To anchor this corporate growth, Anthropic launched its Claude Enterprise Compliance API, introducing 28 security integrations with major providers like Okta, CrowdStrike, and Microsoft Purview. The API securely channels chat text, files, and DLP telemetry straight into existing enterprise SIEM dashboards, solving the regulatory bottlenecks that typically block large-scale AI deployment.

However, massive demand for prolonged agentic workflows has triggered critical capacity shortages. As a response, Anthropic is now spending $1.25 billion a month ($15 billion annually) through May 2029 to rent 300 megawatts of compute from xAI’s Colossus facility, following xAI's recent merger into SpaceX, while simultaneously negotiating early access to Microsoft’s Maia 200 inference chips. Read more.

Index by Parallel tracks exactly how much AI agents rely on your content, helping creators monetize citations and analyze distribution.

Tycoon AI coordinates a multi-agent workforce under an autonomous AI CEO, letting solo founders scale full-stack corporate operations without coding.

VenturusAI reads your quick business idea and automatically runs complex frameworks like SWOT and PESTEL to check its market viability.

Mintlify Workflows connects your codebase directly to your docs, automatically updating knowledge bases and changelogs whenever your product changes.

InstaVM spins up secure, kernel-isolated Linux microVMs in milliseconds, giving your AI agents a safe playground to run code and store data.

Your feedback is valuable. Respond to this email and tell us how you think we could add more value to this newsletter.

Interested in reaching smart readers like you? To become an AI Breakfast sponsor, reply to this email or DM us on X!

Thinking of starting your own newsletter? AI Breakfast readers who sign up with Beehiiv receive a 14-day free trial and 20% off for 3 months.