How I AI: Nick Baumann's 3 Advanced ChatGPT Workflows for Voice Automation, Site Building, and Video Editing

Welcome back to How I AI! I'm Claire Vo, and I'm on a mission to help you build smarter with the latest in AI. I have to admit, I might be the biggest fan of Codex in the world. It’s my daily driver, so I was incredibly excited to sit down with Nick Baumann from OpenAI. Even as a power user, it's hard to keep up with the constant stream of new features, and this episode was a real eye-opener.

Nick is on the Codex team and lives on the cutting edge of what's possible. He walked me through some of the most powerful, and often underused, capabilities of ChatGPT and Codex. We're moving far beyond the simple chatbot interface. We're talking about a future where your AI is a true assistant—one that can see your screen, manage multiple complex tasks, build websites on command, and even edit your video content. It’s about creating systems for high-bandwidth delegation, and Nick showed me exactly how the pros do it.

In this post, we'll break down three specific, actionable workflows Nick shared. You'll learn how he uses ChatGPT Voice to manage his life logistics, how to build and deploy a shareable website in minutes using ChatGPT Sites, and my personal favorite, a brilliant workflow for editing creator content that completely removes the usual tedium. Get ready to have your mind blown.

Workflow 1: Automating Life's Logistics with ChatGPT Voice

We started with a workflow that feels like it’s straight out of science fiction but is available today. Nick showed me how he uses ChatGPT Voice not just for simple queries, but as a high-level orchestrator for his life. The problem it solves is universal: the friction of managing logistical tasks like booking travel and filing expenses. Voice turns this into a simple, high-bandwidth delegation experience.

Step 1: Triggering Voice and Providing Context

It starts with a simple hotkey that brings up the ChatGPT Voice interface, no matter what application you're in. The key here is that the AI can see your screen. Nick had a Slack message open about a team offsite in Paris. Without any copy-pasting, he could just start talking to the AI about it.

The AI takes what Nick calls an "app shot" – essentially a smart screenshot loaded with context that the model can immediately understand and act upon.

Step 2: Delegating Travel Booking

With the context of the offsite understood, Nick moved on to delegation. He gave a natural language command, just as if he were talking to a human assistant.

I'm hoping you can help me book my flights and my hotel for the Paris offsite. I'm also thinking about sticking around through the weekend and maybe a couple days after. Could you maybe like check my calendar, see what my availability's like, and see if it's, if it's feasible for me to stick around?

This single command kicked off a whole chain of actions. The AI confirmed it would check his calendar, look at travel options using its plugins (like Navan), and plan around his request to extend the trip. It even handled the back-and-forth, suggesting the best return date based on his calendar conflicts.

Step 3: Spinning Up a Parallel Task for Expensing

While the travel task was running in the background, Nick demonstrated how the AI can manage multiple, independent threads. He kicked off a completely separate task with another voice command:

While I've got you, I've got an email, I think. There should be some Amazon receipt for a mic I just bought recently. It's for DX uses. I'm hoping you can help me expense it. Could you spin up another thread to handle taking care of that expense report for me?

The AI confirmed it would start a new task, find the Amazon receipt, and stage the expense report in Navan, pausing for his final confirmation before submitting. This is the magic of the new architecture: you can have multiple agents working on your behalf simultaneously, all initiated and managed through a single, seamless voice conversation.

Workflow 2: Instantly Building and Deploying a Shareable Website

Next, Nick showed me a feature I think most people have no idea exists: ChatGPT Sites. You can now prompt the AI to build and deploy a live, data-backed website on a chatgpt.site domain. This isn't just a static page; Nick explained it supports a SQL database, S3 file storage, and environment variables. It’s a complete, self-contained web application builder.

To show me how it works, we decided to build a site live on the podcast.

Step 1: The Initial Idea and Prompt

The idea was to create a site that aggregates all the best tips and workflows from the How I AI podcast, making them easily searchable. I gave Nick the requirements, and he translated them into a single, detailed prompt.

I'm here right now with Claire Vo. She hosts the How I AI podcast, and we're hoping that we can build a site that takes all the best tips from the videos on her YouTube channel and makes it so you can like easily filter through them and have quick links to like different timestamps on her videos. Make it match the How I AI branding. We use good old AI blurple, purple, and black and white. And let's have it, if possible, categorized by tool... and also function... Otherwise, I'm excited to see what you come up with.

Step 2: AI-Powered Development

With that prompt, the AI agent went to work. It began designing the database schema, structuring the front end with the requested branding, and figuring out how to pull in and categorize the data from my YouTube channel. This entire development process, which would normally take a team of engineers days or weeks, was fully automated.

Step 3: Reviewing and Publishing the Site

In a surprisingly short amount of time, the site was ready. It correctly used the purple, black, and white branding and had started pulling in episodes, creating a filterable gallery. Nick then showed me how easy it is to manage access.

You can deploy the site publicly to the entire world, or you can restrict access to a list of specific email addresses. Authentication is handled seamlessly through a user's ChatGPT login. This is perfect for building internal tools, private dashboards, or MVPs that you only want to share with a select group of users.

Workflow 3: AI-Powered Video Editing for Creator Content

As a creator, the last workflow Nick demonstrated was my absolute favorite. I spend so much time editing videos for the podcast, and the process is incredibly tedious. Nick showed me how he automates the entire thing, from selecting the best takes to cutting the final video and even redacting sensitive information.

Step 1: Ingesting Raw Footage

Nick started by simply dragging a folder of raw, unedited video clips directly into the ChatGPT Work interface. These were from a recent shoot where he filmed himself in a park, with multiple takes for each part of the video.

Step 2: Prompting the Edit with a Custom Plugin

He then dictated a prompt explaining the story he wanted to tell, using a custom UGC Video plugin he built. This plugin contains his preferences for video formats (9x16, 4x5), caption styles, and other stylistic elements.

I recorded a video for the record and replay feature. It starts with me just explaining that I'm apartment hunting... and then I walk through and show the feature in real life... Can you go through these clips first, like pull the transcripts, find the best takes, and then kind of piece this together into a UGC-style video?

Step 3: The AI Editing Process

This is where the real magic happens. The AI, which Nick specified should run on the more powerful Sol Ultra model framework, initiated a complex, multi-agent process:

1. Transcription: It first processed all the clips to get a full transcript.
2. Take Selection: It analyzed the transcripts and his verbal cues (like saying "cut" or "that's a good take") to identify the best segments to use.
3. Visual Analysis: It processed frames from the videos to understand the visual content.
4. Assembly: It pieced together the best takes into a coherent narrative.
5. Automatic Redaction: This part was incredible. In the video, his screen showed some sensitive information (like addresses). The AI automatically identified and added animated blurs that tracked the sensitive content perfectly as he scrolled. It even checked its own work, re-blurring frames until the redaction was perfect.

This workflow eliminates some of the most embarrassing and time-consuming parts of being a creator. I love the idea of offloading the self-critique of watching my own takes and the painstaking process of blurring sensitive information to a neutral third party.

A New Era of AI Assistance

These three workflows represent a significant shift in how we interact with AI. We're moving away from a simple request-response model and toward a relationship of delegation and partnership. Whether it's managing your calendar through voice, building a functional website with a single prompt, or automating a complex creative task like video editing, the tools are here to turn AI into a true assistant that works for you in the background.

What Nick Baumann demonstrated isn't just a collection of cool features; it's a glimpse into a more efficient and creative future. The barrier to entry for these powerful capabilities is getting lower every day, empowering not just engineers but everyone to automate the tedious parts of their work and life. I hope you'll take these examples and start exploring what you can build.

Thanks to Our Sponsors

A huge thank you to our sponsors for this episode, who are making it possible for more people to build with AI.

• [Bolt.new](http://bolt.new/howiai): The AI app builder for people who have ideas and want to ship them. You describe what you wanna build—a startup MVP, a landing page, an internal tool, a side project—and Bolt generates production-ready code in minutes. You just need an idea and a weekend. Check it out at bolt.new/howiai.
• [Hyperagent](https://hyperagent.com/howiai): The platform for deploying always-on agents that actually run your business. With Hyperagent, you build agents in the cloud and deploy them where your work already happens, like Slack, Telegram, or email. These aren't chatbots waiting for a perfect prompt. They're proactive, learning your preferences, retaining your playbooks, and getting better with every run. How I AI listeners get one thousand dollars in free inference to start building. Claim yours at hyperagent.com/howiai.

Episode Links

• YouTube: https://youtu.be/EgrI2rUmM48
• Spotify: https://open.spotify.com/episode/6HlPJh4X8tlCUjEMonOpx4?si=Orgg80JpTb-w-hdcEAA8Xw