Good morning. It’s Friday, February 6th.

On this day in tech history: In 1996, Juniper Networks was founded by Pradeep Sindhu, focusing on high-performance routers with custom ASICs for packet forwarding. This disrupted Cisco's dominance, enabling scalable IP networks crucial for distributed AI workloads.

In today’s email:

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

The AI coding wars reached a fever pitch this week as OpenAI and Anthropic launched rival models minutes apart. OpenAI released GPT-5.3-Codex, a specialized model that didn't just learn to code, it helped build itself.

In a new milestone for "recursive" AI development, OpenAI engineers used early versions of the model to debug its own training runs, manage the infrastructure it lives on, and evaluate its own performance tests. This self-correction cycle has resulted in a model that is 25% faster than its predecessor and far more capable of handling complex, real-world technical tasks.

On key benchmarks, GPT-5.3-Codex achieved a dominant 77.3% score on Terminal-Bench 2.0, outperforming Anthropic’s newly released Claude Opus 4.6 by roughly 12 points. Beyond just writing code, the model is designed to act as a "computer operator," capable of navigating desktop environments and managing entire software deployments autonomously.

As the underlying models get smarter, OpenAI is also changing how companies use them. The company introduced Frontier, a platform that converts AI from a tool into a "digital employee."

The power of these autonomous systems was further proven in the lab. In a partnership with Ginkgo Bioworks, a GPT-5-based system autonomously designed and executed 36,000 protein synthesis experiments over six months. By handling the "thinking" (hypothesis and data analysis) while robots handled the "doing," the AI reduced the cost of producing specific biological materials by nearly 60%.

As the technical capabilities grow, so does the corporate friction:

Minutes after OpenAI’s latest move, Anthropic has fired back with Claude Opus 4.6, a major upgrade designed to turn AI from a single assistant into a coordinated workforce. This version doesn't just think faster, it thinks bigger and works in groups.

The headline feature is "Agent Teams." Instead of one AI trying to solve a massive problem alone, Claude can now deploy multiple specialized agents to work on different parts of a project simultaneously.

To prove it works, Anthropic’s researchers tasked a team of 16 Claude agents to build a 100,000-line software compiler from scratch. Working autonomously over 2,000 sessions, the agents coordinated through a lead "manager" who assigned tasks like coding, testing, and documentation. While the project cost roughly $20,000 in API fees, it demonstrated that AI can now manage large-scale engineering projects with almost no human intervention.

A 1-Million Token Memory. Context is everything in AI, and Opus 4.6 has expanded its "memory" to a massive 1 million tokens.

Winning the Corporate Office. Anthropic is aggressively targeting professionals in finance and law. On enterprise benchmarks, Claude Opus 4.6 significantly outscored OpenAI’s GPT-5.2, proving its strength in high-level knowledge work. New features include:

As these systems gain the power to act on their own, safety is a top priority. Anthropic’s tests show that while the model is remarkably good at solving novel problems, it maintains strict security protocols. The company has implemented new cybersecurity probes to ensure that these "agent teams" aren't easily tricked into harmful behavior, keeping the focus on productivity. Read more.

📄 AOrchestra: Automating Sub-Agent Creation for Agentic Orchestration

📄 ERNIE 5.0 Technical Report ​​

📄 SWE-Universe: Scale Real-World Verifiable Environments to Millions

📄 FSVideo: Fast Speed Video Diffusion Model in a Highly-Compressed Latent Space

📄 X-Reasoner: Towards Generalizable Reasoning Across Modalities and Domains

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!