Last week, voice mode was blowing up：https://x.com/danshipper/status/2082613916706693560 Every’s Slack.
Powered by GPT-Live：https://openai.com/index/introducing-gpt-live/ , OpenAI’s new voice model, the feature lets you have natural conversations with ChatGPT：https://every.to/context-window/the-urge-to-merge-chatgpt-and-codex , complete with interruptions, follow-up questions, and redirections. Within the ChatGPT desktop app, voice mode can find the right task or thread based on spoken context, kick off new threads, check on existing work, and send more complex tasks to GPT-5.5：https://every.to/vibe-check/gpt-5-5 .
After Dan took to X to evangelize voice mode’s powers：https://x.com/danshipper/status/2082839820657623243 for writing and revising an essay, the team put the feature through its paces. We used it to fix user-reported bugs, draft article outlines：https://x.com/kplikethebird/status/2082856545499365563 , do meal prep, draw connections：https://x.com/leeknowlton/status/2082389588237303891 between what we were reading and what we were building, book airline tickets, and orchestrate agents while cooking.
What works: There’s a lot to love about voice mode, which allows you to get work done without having to sit at a keyboard.
One of its biggest strengths is that it lets you read and ask questions aloud or connect what you’re reading to another file or project. Engineer Lee Knowlton uploaded a PDF of Designing Data-Intensive Applications：https://www.oreilly.com/library/view/designing-data-intensive-applications/9781491903063/ , a book about building large-scale data systems, while voice mode had access to the codebase he was working on. As a result, he could keep reading：https://x.com/leeknowlton/status/2082389588237303891 while asking questions aloud, exploring unfamiliar ideas, and connecting the book’s insights to his own code. The result was a more fluid way to learn.
“Shifting from text to voice is different from shifting from text to text for me,” he says. “Reading something and then having a conversation, or asking a quick question, is different from typing something and then having to parse more text.”
What could be better: During a walk, COO Brandon Gell：https://every.to/@brandon_5263 found that the mobile app’s voice mode could read a thread’s visible history but didn’t have access to important context outside the thread. Currently, voice can control local Codex work through Remote：https://learn.chatgpt.com/docs/remote-connections connections—but only while the host computer is awake, online, and running the desktop app. Without that connection, voice can’t access the host’s Codex：https://every.to/podcast/how-openai-s-codex-team-uses-their-coding-agent projects, files, or tools.
Further complicating matters, the mobile app also has an “ordinary voice mode,” which can use the current cloud conversation but not the local context available through Remote. Are you confused? We’re confused.
The model’s ability to distinguish between speech intended for it and ambient conversation was also inconsistent. Lee found it good at filtering out exchanges with his wife, while engineer Tyler Nishida had the exact opposite experience. And the lag time can make it hard to use as a writing or editing partner (I found voice mode impressive but functionally too laggy to help me write this piece, for example). Finally, although GPT‑Live can delegate complex tasks to a frontier model in the background, some responses still felt shallow compared with responses from a text chat set to GPT-5.6 Sol：https://every.to/vibe-check/gpt-5-6-sol .
Final verdict: Voice mode is, as Dan puts it, “a whole new world”—one in which you can direct agents away from a computer. But there are kinks to work out.
“It’s both not quite there yet and obviously the future,” Lee says. “A week ago, I couldn’t imagine a version of this that was really good, and now I can.”
Customer service volume doesn’t scale with headcount—ElevenAgents does. The platform deploys voice and chat agents that handle billing questions, support tickets, and outreach around the clock across 70+ languages, integrating with Salesforce, Zendesk, and the tools your team already runs. Faster resolutions mean fewer abandoned carts and higher LTV.
Sarah Tavel has spent her career studying consumer technology cycles, and she thinks she’s spotted the next one. A former Pinterest product manager and current Benchmark partner, she’s betting the next wave of AI products won’t just be smarter, they’ll be social. On this week’s AI & I：https://every.to/on-every/introducing-ai-i , we’re revisiting our April 2025 conversation with Sarah, who argues that even power users are still using AI products like ChatGPT in a rudimentary way. But the gap isn’t the models: It’s that nobody has built a way for users to learn from each other. Sarah thinks whoever captures and shares that knowledge will create the next big product.
Watch on X：https://x.com/every/status/2085059043970650326 or YouTube：https://youtu.be/dlI-5W7d7uU , or listen on Spotify：https://open.spotify.com/episode/1nDCkDLbKYuj4mdJlAvPcY?si=_KylC4uSREitlsLEoczWFw or Apple Podcasts：https://podcasts.apple.com/us/podcast/why-the-next-hit-ai-product-will-be-social/id1719789201?i=1000780083451 . You can also read the transcript：https://every.to/podcast/transcript-c09360f3-efda-4688-952d-203b9f5f4315 .
This episode is a must listen for anyone who wants to understand why the biggest AI product hasn’t been built yet—and what it might take to build it.
Miss an episode? Catch up on Dan’s recent conversations with Anthropic head of product Mike Krieger：https://www.youtube.com/watch?v=KRv9GpJYrUA ; the team that built Claude Code：https://every.to/source-code/claude-code-for-product-managers , Cat Wu：https://every.to/podcast/how-to-use-claude-code-like-the-people-who-built-it and ：https://every.to/podcast/how-to-use-claude-code-like-the-people-who-built-it Boris Cherny：https://every.to/podcast/how-to-use-claude-code-like-the-people-who-built-it ; the team that built Codex, Thibault Sottiaux：https://every.to/podcast/how-openai-s-codex-team-uses-their-coding-agent and ：https://every.to/podcast/how-openai-s-codex-team-uses-their-coding-agent Andrew Ambrosino：https://every.to/podcast/how-openai-s-codex-team-uses-their-coding-agent ; Vercel cofounder Guillermo Rauch：https://every.to/podcast/vercel-s-guillermo-rauch-on-what-comes-after-coding ; podcaster Dwarkesh Patel：https://every.to/podcast/dwarkesh-patel-s-quest-to-learn-everything ; and others to learn how they use AI to think, create, and relate.— Miriam Partington：https://www.linkedin.com/in/miriam-partington-499b71149/
Lee’s workflow shows how voice mode changes the way we interact with agents...
The essential toolkit for those shaping the future
"This might be the best value you can get from an AI subscription."
Join 100,000+ leaders, builders, and innovators
Already have an account? Sign in：/login.
Daily insights from AI pioneers + early access to powerful AI tools
Sparkle : Organize your Mac with AI
Cora : The most human way to do email
Spiral : Repurpose your content endlessly
Monologue : Effortless voice dictation for your Mac
Scan to open this article