How this OpenAI engineer uses Codex + ChatGPT Work to automate everything
Nick Bowman from OpenAI demonstrates advanced use cases of ChatGPT and Codex, including voice-orchestrated multi-threaded task management, AI-powered website creation via ChatGPT artifacts, and automated video editing for content creation. He emphasizes how these tools are becoming accessible to non-technical users through mobile and voice interfaces.
Summary
In this episode of 'How I AI,' host Clarvo interviews Nick Bowman, an OpenAI engineer, about cutting-edge features in ChatGPT and Codex that most users don't yet know about. The conversation covers three major use cases:
First, ChatGPT voice functionality (released the day before this recording) enables high-bandwidth delegation of complex tasks. Nick demonstrates booking flights and hotels for a Paris offsite while simultaneously creating an expense report—all managed through voice commands with multiple concurrent threads running in the background. The voice feature can see the user's screen via appshots (smart screenshots), access plugins and browser capabilities, and manage up to five or six threads simultaneously. Unlike typical voice AI that uses weaker models, this implementation uses ChatGPT's full intelligence, allowing users to delegate with confidence.
Second, ChatGPT artifacts deployed on custom domains (chatgpt.site) allow users to build and deploy functional websites with SQL databases, S3 storage, and environmental variables. Clarvo and Nick build a live website aggregating tips from How I AI episodes, filterable by tool and function. These sites support access controls (public, private, or email-restricted), can be edited from mobile, and are deployed instantly—representing a democratization of web app creation.
Third, Nick showcases video editing automation via Codex, where he dumps 50-60 raw video clips into ChatGPT with voice notes about desired edits. The system transcribes clips, processes image frames, identifies the best takes based on verbal cues ("cut," "bad take"), and assembles finished UGC-style videos in multiple formats (9x6, 4x5, organic vertical). The system even automatically redacts sensitive information like addresses and API keys with frame-following blur effects. This workflow reduces the barrier to content creation from hours of manual editing to simple recording and voice direction.
Key themes throughout include: the shift from blank-page syndrome through voice as a 'yapper's API,' latency versus intelligence tradeoffs in voice (Nick argues latency matters more when good tools are available for delegation), the challenge of using voice in open offices (requiring meeting pods and whispered dictation), and the importance of discoverability for new features. Nick emphasizes that most people don't know about meta-capabilities like thread creation and forking, which allow AI to manage context intelligently.
Key Insights
- ChatGPT voice represents a new primitive where full-intelligence models can manage multiple concurrent threads (5-6 threads) simultaneously with delegation capabilities, unlike typical voice AI that trades off capability for latency
- Nick argues that when delegation tools are available, latency matters more than raw intelligence for voice experiences, as users can tolerate thinking time if the AI effectively offloads work to available plugins and services
- ChatGPT mobile and work tab features enable non-technical users to set up always-on agents monitoring email and finances, representing a democratization of agent capabilities previously limited to developers and AI tinkerers
- Video editing automation using Codex can automatically handle sensitive redaction tasks by checking frames iteratively until good results are achieved, solving a previously tedious manual problem for content creators
- Nick observes that people in OpenAI offices are increasingly using private meeting pods and pods specifically to use voice dictation features, indicating organic workplace adaptation to voice-based AI workflows despite infrastructure challenges
Topics
Transcript
[0:00] If we go back to, you know, 2022, 2023 around the release of chatbt, I think a lot of people felt the AI then a chatbot that I say things to and intelligently says things back to me. I think with the advent of coding agents and like the coding agent harness in particular, now this AI is going out and like reading files and creating files, like running commands and this is crazy. I sell it one thing and then it figures out all the other things on it own. >> I haven't gone deep with voice. I want to see it from the pro. How should we be using it? What's cool about it? >> So, I'm…
Full transcript available for MurmurCast members
Sign Up to AccessMore from How I AI
How (and why) to build a personal API
The conversation explores the concept of personal APIs that store individual preferences like favorite restaurants and coffee orders, enabling others to make thoughtful gestures without asking. The discussion evolves to envision AI agents autonomously accessing these APIs to perform personalized tasks like making reservations.
Hacking an old pager with AI
A developer describes setting up an old pager as a Twitter notification device using modern cloud services. The system routes Twitter API events through Cloudflare Workers, Resend email service, and Gmail to finally deliver notifications to the pager's email address.
Opus 5: too timid?
The speaker criticizes Opus 5 for being excessively cautious and indecisive, constantly asking for human permission or approval rather than taking action. Even when faced with simple tasks like fixing a one-line merge conflict, Opus 5 deferred to the original branch owner's preferences, requiring the speaker to repeatedly push it to make autonomous decisions.
GPT-5.6's video editing via Codex is genuinely one of my favorite new workflows
A product manager describes using GPT-5.6 with Codex to automate video editing for social media clips. By simply dragging a file and providing natural language instructions, the AI generated five polished, fast-paced hype videos from a long conference talk recording, dramatically reducing the time-intensive manual clipping process.
Theoretically Intelligent vs. Practically Effective: Why GPT-5.6 Sol Beats Fable for Product Work
An executive contrasts two AI models (Fable and Soul/GPT-5.6 Sol), arguing that Soul is superior for product work because it prioritizes practical effectiveness over theoretical intelligence. The speaker values the ability to ship products to customers and understand end-user goals over theoretical sophistication.