• The AI Operator Weekly
  • Posts
  • AI is literally saving lives...and OpenAI announces unbelievably realistic voice agents

AI is literally saving lives...and OpenAI announces unbelievably realistic voice agents

AI medical breakthroughs, hyper-realistic voice conversations, and a guide to creating forms in minutes with AI.

AI Operators!

I’m preparing to jet off to Berlin on Thursday where my two passions – AI and music – are about to collide for one exciting night.

Our first ever client, nonplusultra, are having their 10-year anniversary party and the team have asked me to be their official “DJ and AI guy”, which honestly made my week.

Shout out to the nonplusultra team who’ve been with us since day 1 and hold a special place in my heart!

Speaking of hearts, I can’t wait to share this week’s inspiring news about a new AI device that could save millions from heart disease.

Let’s get into it!

Today at a glance:

  • In the news: Best-ever voice agents from OpenAI, a Claude Chrome extension in the works, and the AI-powered stethoscope saving lives.

  • Guide: Create custom AI interfaces without coding using MindStudio's new feature

  • Media: ChatGPT's new project memory feature keeps client information contained

  • Tool: OpenAI Codex works as your AI coding teammate across multiple environments

News: 

OpenAI release most advanced voice model, and API to build voice agents

OpenAI has made their Realtime API generally available for building production voice agents, with new features including remote server support, image inputs, and phone integration.

They've also released gpt-realtime, an advanced speech-to-speech model that's much better at following instructions, using tools precisely, and speaking naturally. It can switch languages mid-sentence, and handle complex tasks like reading scripts verbatim. We're finally getting truly capable AI voice agents with natural, expressive speech.

Anthropic piloting a Claude Chrome extension

Anthropic has announced Claude for Chrome, a browser extension (currently on pilot) that lets Claude take actions directly in your browser.

It’s an awesome next step after Claude's integration with calendars, documents, and other software, and will be a massive productivity boost for browser-based tasks.

There are some security concerns—AI agents can sometimes be tricked into harmful actions by malicious website content. But Anthropic is carefully testing with 1,000 users and working to boost the tool’s security.

You can join the waitlist to try it out!

The AI stethoscope that detects heart problems in seconds

Doctors at Imperial College London developed an AI-powered stethoscope that can diagnose heart failure, heart valve disease, and abnormal heart rhythms in just 15 seconds, with patients being up to three times more likely to receive accurate early diagnoses compared to traditional methods.

This is exactly the kind of news that brightens my day - AI being integrated into medical devices to help people with heart problems get diagnosed sooner.

(I actually predicted something like this in an AI Operator blog post of 2025 trends and predictions!)

One Guide :

Create Custom Forms Without Coding: MindStudio's New Feature

With MindStudio's new custom form designer, you can create professional-looking forms and interfaces through simple conversation. This means you can transform basic inputs into engaging, branded forms without writing a single line of code.

Step 1: Open up the builder

In MindStudio, navigate to any User Input block and scroll down to the "form" section. Change the setting from "native" to "custom" and click "configure form" to open the development environment.

Step 2: Start a conversation

The magic of this tool is what they call "vibe coding" — essentially telling the system what you want in plain English. Begin by describing your ideal form in the chat window, such as "I need a form that collects employee information with a friendly, modern look."

Step 3: Refine through dialogue

The system immediately shows you a preview of what it’s creating based on your description. Don’t like something? Simply chat more specific instructions like, "Make the background lighter" or "Add our company logo at the top."

Step 4: Watch your page evolve in real-time

As you continue the conversation, the form updates instantly with each new instruction. This immediate feedback loop makes the design process intuitive and efficient, saving hours compared to traditional development.

Step 5: Finalize your creation

Once you’re satisfied with how everything looks and functions, click “compile.” The system handles all the technical work behind the scenes, packaging everything up and making it ready to use in your agent.

Step 6: Implement and enjoy

After compilation, refresh your agent to see the new form in action. Your users will now interact with your custom-designed form instead of the default version.

Final Step: Put it to work in your business

This feature is perfect for creating professional onboarding forms, intuitive file upload systems, multi-step forms with smart validation, or even fun, interactive designs for creative applications like story generators.

You can update and refine these forms anytime through the same process.

One Media:
ChatGPT's New Project Memory Feature

I just dropped a new video covering ChatGPT's project memory feature, and it’s going to be a productivity boost for businesses.

The feature creates separate memory compartments for each project, so your client conversations stay contained within their respective projects and don't bleed into other work.

The update solves two major problems at once, which explain in detail in the video:

One Tool:

OpenAI Codex: Your AI Coding Teammate

I've been putting OpenAI's Codex through its paces, and I'm impressed by the latest updates.

Codex is a coding agent that can read, modify, and run code directly on your local machine. What makes it stand out is its ability to work everywhere you code - your terminal, IDE (VSCode, Cursor, Windsurf), in the cloud, and even on your phone through the ChatGPT mobile app.

You can start a task on your laptop, delegate complex work to run in Codex's cloud sandbox, review changes on your phone, and merge everything back without losing context.

Key capabilities:

  • Multi-environment coding - Works in terminal, IDE extensions, and cloud sandboxes

  • Automated code reviews - Integrates directly with GitHub for PR reviews and task management

  • Mobile coding - Kick off tasks and review code changes from your phone

  • Isolated execution - Runs code safely in sandboxes before you merge changes

For non-technical founders, this could be a game-changer for understanding what your development team is building. Instead of waiting for explanations, you can ask Codex to walk you through code changes in plain English.

For technical founders wearing multiple hats, Codex can help you quickly prototype features or squash critical bugs without context-switching between different tools.

I'm currently testing Codex against Claude Code to see which one fits better into my workflow. Both are impressive, but I need another week of real-world usage before I can give you my honest comparison.

Check out OpenAI Codex here and see if it fits into your workflow.

TL;DR:

  • I'm flying to Berlin to DJ at the 10-year anniversary party of our first ever client, nonplusultra.

  • Exciting releases from OpenAI and Anthropic, and AI starts saving lives in the medical world.

  • MindStudio's new "vibe coding" feature lets you create custom AI interfaces through simple conversation, no coding required.

  • My new video explains how ChatGPT's project memory feature keeps information separate, improving focus and privacy.

  • OpenAI’s Codex can be your coding teammate, so technical and non-technical founders can manage code more effectively.