How I AIHow this OpenAI engineer uses Codex + ChatGPT Work to automate everything
At a glance
WHAT IT’S REALLY ABOUT
OpenAI engineer demos voice-driven Codex agents automating work and content
- Codex Voice acts as a screen-aware assistant that can spin up and coordinate multiple parallel threads to complete real tasks like booking travel and preparing expense reports without immediate supervision.
- ChatGPT Work on mobile extends agentic workflows with connectors and proactive “Heartbeats” that monitor sources like email, calendar, and finances and notify you when something needs attention.
- The conversation highlights under-discovered capabilities such as thread forking, cross-thread messaging, AppShots (context-rich screenshots), and using an existing browser session for authenticated flows.
- They build and deploy a ChatGPT Site that can be shared publicly or restricted by email, positioning “Sites” as a lightweight way to publish databases/files-backed web artifacts.
- A repeatable creator workflow emerges: dump many raw clips into Codex, use a custom plugin to enforce brand/safe-area rules, auto-select best takes via transcripts, and even perform moving redactions/blur checks.
IDEAS WORTH REMEMBERING
5 ideasTreat Voice as an orchestration layer, not just dictation.
In the demo, Voice isn’t merely transcribing—it creates new threads, updates running tasks, and uses screen context to delegate multiple workflows (travel + expensing) concurrently.
Use threads to manage context and parallelize work intentionally.
Codex can start/fork threads and have them work in the background, which reduces context overload in one chat and mirrors how a human assistant would split responsibilities.
Prefer ‘use my existing browser session’ for real-world web automation.
When tasks require logged-in access (e.g., Navan, Gmail), instructing the agent to use the existing Chrome session avoids re-auth friction and keeps the workflow practical.
Heartbeats turn ChatGPT Work into an always-on monitor for life admin.
By connecting email/calendar/finance tools and running recurring checks, you get proactive notifications (packages, suspicious charges) instead of needing to remember to look.
AppShots are a fast path to ‘agent understands what I’m seeing.’
A context-rich screenshot of the active window lets the system immediately reason about what’s on-screen (like a Slack itinerary) and act without manual copy/paste.
WORDS WORTH SAVING
5 quotesI tell it one thing, and then it figures out all these other things on its own.
— Nick Baumann
I think with the advent of code agents and like the code agent harness in particular, now this AI is going out and like reading files and creating files and like running commands, and it... This is crazy.
— Nick Baumann
What I like about Voice is I do think sometimes people with AI really struggle still with, like, blank chat window syndrome.
— Claire Vo
If anybody has been lucky enough to have an assistant, like, this is how it gets, it gets done, and it is just, like, such a nice delegation experience to be like, "Hey, can you take care of this? Hey, can you fix this? Tell me what's going on."
— Claire Vo
There's gonna be delay. Um, a- and the more we shrink that, we can kind of drop those assumptions and new products will, you know, will emerge.
— Nick Baumann
High quality AI-generated summary created from speaker-labeled transcript.
