🚨 GPT 5.3 Codex vs Opus 4.6

Good Morning! GPT-5.3-Codex and Claude Opus 4.6 dropped on the same day, and instead of one clear winner, each dominates a different battlefield. Plus, you’ll learn how to use Claude Cowork better than 99% of people step by step.

Plus, in today’s AI newsletter:

• Anthropic Launches Claude Opus 4.6

• OpenAI Releases GPT-5.3-Codex

• OpenAI Launches Frontier for Managing AI Agents

• How to Use Claude Cowork Better Than 99% of People

• 4 new AI tools worth trying

AI MODELS

🧑‍💻 Anthropic Launches Claude Opus 4.6

Anthropic unveiled Claude Opus 4.6, its most capable model yet, built for sustained, real-world work across coding, research, and enterprise workflows. The company says it marks a shift from quick AI tasks to delegating serious professional work.

• Stronger at coding, planning, code review, and debugging across large codebases

• Excels at long-running tasks, research, and financial analysis

• Now ranks #1 on the Finance Agent benchmark

• Available via claude.ai, API, and all major cloud platforms

Anthropic is signaling a move from AI as a helper to AI as a coworker. As enterprises already make up ~80% of its business, Claude Opus 4.6 could accelerate AI adoption where reliability, depth, and long-term task ownership actually matter.

AI MODELS

🧑‍💻 OpenAI Releases GPT-5.3-Codex

GPT-5.3-Codex is OpenAI’s newest coding-focused model, combining GPT-5.2-Codex’s programming strength with GPT-5.2’s broader reasoning and knowledge, while running significantly faster.

• Runs ~25% faster and uses fewer tokens than earlier Codex models

• Beats Anthropic’s Opus 4.6 by 12 points on Terminal-Bench 2.0

• Scores 64.7% on OSWorld, far ahead of GPT-5.2-Codex’s 38.2%

• Was used internally to find bugs, manage deployment, and evaluate its own training

GPT-5.3-Codex shows how AI is becoming a force multiplier for building AI itself. Stronger coding + agentic computer use means faster software development, and raises the stakes on safety, reflected in its new “High” cybersecurity risk classification.

AI NEWS

🧑‍💼 OpenAI Launches Frontier for Managing AI Agents

OpenAI Frontier is a unified control layer that helps businesses build, deploy, and manage AI agents, including agents not made by OpenAI. It’s built to solve the chaos of fragmented tools, siloed data, and ungoverned agents inside enterprises.

• Acts as an “agent interface” that gives AI agents shared context, permissions, and boundaries

• Lets companies onboard and manage AI agents like employees, with feedback and learning loops

• Supports use in sensitive and regulated environments with clear controls

• Already piloted by companies like Intuit, State Farm, Uber, and Thermo Fisher

As companies start “hiring” AI agents at scale, governance becomes the bottleneck. Frontier signals OpenAI’s push to own the enterprise layer where AI agents are managed, trusted, and integrated into real business workflows.

HOW TO AI

💻 How to use Claude Cowork better than 99% of people (Full Guide)

In this tutorial, you’ll learn how to use Claude Co-Work to move beyond chat and turn AI into a true daily work partner.

🧰 Who is This For

• Professionals who want AI to help with real day-to-day work, not just ideas

• Creators and operators with repeatable workflows they want to speed up

• Builders who are tired of juggling prompts, projects, and tools

• Anyone who wants an AI assistant that plans, executes, and iterates with them

STEP 1: Access the Co-Work Desktop Workspace

To get started, you’ll need to install and open Claude desktop app.

Once your desktop app is updated, open it and look at the top navigation. You’ll see a second tab labeled Co-Work, positioned next to the code-focused workspace. Click into it to enter the main co-work environment. This is where file access, software connections, skills, and execution all live together.

STEP 2: Give It Access to Your Files and Folders

One of the first things that makes co-work different is direct file access. You can grant it permission to specific folders on your computer, such as Downloads, project directories, or business documents.

After selecting a folder, you can ask it to take real actions. For example, you can tell it to organize your Downloads folder by file type. Before acting, it will pause to plan, ask clarifying questions, and show you exactly what it’s about to do. Once confirmed, it executes the plan step by step and completes the task automatically.

Beyond organization, this file access is powerful for context. You can store things like your tone of voice, ICP, brand docs, or strategy files in folders and let the AI reference them whenever it works on new tasks.

STEP 3: Connect Your Software Stack

Next, you can connect your tools so the AI can read and update real data. Inside settings, open the Connectors section. Here you’ll find built-in integrations for popular tools. Once connected, the AI can fetch information and take actions directly.

If a tool isn’t listed, you can manually connect it using MCPs. This involves adding a small configuration entry to a desktop config file. Once set up, the AI treats that software like a native integration.

If no API or MCP exists, you still have a fallback: browser use. In this mode, the AI opens and operates your browser just like a human would, clicking, scrolling, researching, and extracting information, while you continue working on other tasks in parallel.

STEP 4: Build and Use Skills to Automate Real Workflows

This is where co-work becomes a game changer. Skills are reusable instructions that bundle process, context, and knowledge sources into one callable unit. Think of them as smarter system prompts that only activate when needed and can be combined in the same task.

You can start with built-in skills found in the settings under example skills. You can also import skills created by others from public libraries. Once added, you trigger them simply by asking the AI to use that skill.

The most powerful option, however, is creating your own skills. You do this by walking through a task manually once, step by step, with the AI. For example, you might repurpose a YouTube video into a newsletter by first extracting the transcript, then choosing subject lines, then writing the final draft in your voice. After completing the process once, you ask the AI to save it as a skill.

From that point on, you can repeat the entire workflow with a single instruction. The AI follows the same steps, uses the same context, and even updates connected tools automatically. Over time, you can refine these skills, combine multiple skills in one session, and automate large parts of your daily work without losing flexibility or control.

Anthropic introduces “agent teams” in Claude Code as a research preview to launch multiple agents that work in parallel, and a Claude in PowerPoint plugin.

Apple wound down plans for an AI-based virtual health coach in recent weeks; Eddy Cue has told colleagues that Apple needs to move faster in health.

Tim Cook told Apple staff at an all-hands that he's “deeply distraught” with the US' current approach to immigration and he will press the issue with lawmakers.

Anthropic says Opus 4.6 found 500+ previously unknown high-severity security flaws in open-source libraries with little to no prompting during its testing.

🧑‍💻 Claude Opus 4.6: Built for long, serious work in coding and research

⚙️ GPT-5.3 Codex: Faster coding model with strong reasoning

🚀 Codex App: OpenAI’s Mac app to run and manage AI agents

🧠 OpenClaw: An open-source AI assistant that works inside chat apps

THAT’S IT FOR TODAY

Thanks for making it to the end! I put my heart into every email I send, I hope you are enjoying it. Let me know your thoughts so I can make the next one even better!

See you tomorrow :)

- Dr. Alvaro Cintas