codex
Coverage of codex in the Nexus archive.
- OpenAI debuts ChatGPT Voice so you can have ongoing conversations, ask the AI to complete tasks while hands-free
OpenAI has launched ChatGPT Voice, a hands-free voice feature for the ChatGPT desktop app, enabling users to converse with AI and delegate tasks while working. The technology extends GPT Live voice capabilities from mobile to desktop, targeting software programmers and AI power users through integration with Codex and ChatGPT Work.
- OpenAI's latest proposal shows it's very aware that data centers are unpopular
OpenAI plans to build a $20 billion data center campus near Savannah, Georgia, called Project Camellia, which will require 3.2 gigawatts of power. The company pledged to avoid increasing local electricity or water costs and will provide community benefits and AI credits for Georgia students.
- OpenAI is launching new corporate software that takes it beyond the AI model war
OpenAI launched Presence, a corporate software service designed to integrate AI agents with internal company data and workflows, aiming to automate tasks like customer support and sales. The move positions OpenAI to compete in the enterprise software market beyond AI model sales, as rivals like Anthropic, Google, and Meta intensify competition.
- Cursor, Codex, Gemini CLI, Antigravity hit by sandbox escapes
Researchers exploited sandbox escapes in Cursor, Codex, Gemini CLI, and Antigravity by having AI agents write files executed by trusted host tools. The vulnerabilities led to multiple CVEs, patches, and Google downgrading two Antigravity findings.
- OpenAI has seen a 'resurgence' of interest in secondary markets.
OpenAI has experienced a resurgence in secondary market demand after releasing new models and the AI coding agent Codex, though Anthropic remains the more sought-after company. Investors previously favored Anthropic, but OpenAI's recent advancements have increased its appeal in an illiquid market where shares are traded pre-IPO.
- Codex starts encrypting prompts, uses ciphertext for inference instead
Codex has begun encrypting prompts and now uses ciphertext for inference instead. The article is referenced in a GitHub issue and a Hacker News comment thread.
- My colleague wanted a nerdy, niche tool for his MacBook. Codex whipped it up in minutes.
Stephen Council used OpenAI's Codex tool to create a custom clipboard manager app for his MacBook in under 5 minutes. The app allows storing multiple text snippets across 9 slots with keyboard shortcuts, enhancing productivity without coding.
- OpenAI is making its biggest play for the office
OpenAI is integrating its Codex AI coding tool into the ChatGPT desktop app to create a unified work hub, alongside releasing new GPT-5.6 AI models. The company aims to compete with rivals like Anthropic, Google, and Meta while expanding Codex's use beyond coding to areas like data analysis and research.
- AI makes mistakes, too
Reed experienced unexpected $500 charges from OpenAI's Codex due to a software glitch causing token overuse. He attempted to resolve the issue via Codex and OpenAI's customer service but received no refund, highlighting the financial risks of unmonitored AI usage.
- GPT-5.6 Sol Ultra will be in Codex
GPT-5.6 Sol Ultra is set to be included in Codex. The announcement was shared on Twitter by thsottiaux and discussed on Hacker News with 48 points and 13 comments.
- When I lost access to Fable, I had a backup plan. Now that it's back, I'm even more certain that my fail-safe strategy works.
Anthropic restored access to Fable 5 after the White House blocked foreign access, allowing UK-based business owner Sean McDonnell to continue using the AI model. McDonnell emphasized the importance of backup plans for AI disruptions, using strategies like saving codebase data as Claude Skills and documenting workflows to mitigate risks.
- Ask HN: Is anyone experimenting with different ways of using LLMs for coding?
A user expresses frustration with current LLM tools like Claude Code and Codex for programming, stating they hinder flow state and feel disruptive. They question if alternative approaches beyond the standard prompt-response model are being explored, particularly advocating for a tab model and seeking information on startups or experiments in this space.
- AI coding has split software engineers into tribes. Meet the fans, skeptics — and those in between.
AI coding tools like Anthropic's Claude and OpenAI's Codex are reshaping software engineering, dividing developers into enthusiasts, skeptics, and those with mixed views. A 2025 Stack Overflow survey found 60% of developers hold positive views, though challenges like token budget battles and workplace adjustments persist.
- OpenAI sets up 'warroom' to investigate users burning through Codex credits faster than usual
OpenAI is investigating Codex users hitting usage limits faster than usual due to a fraud-prevention system error. The company reset user caps and is monitoring the situation.
- Show HN: Adrafinil – keep a lid-closed Mac awake only while agents work
Adrafinil is a macOS tool that prevents laptops from sleeping when the lid is closed, specifically while AI agents like Claude Code or Codex are active. It uses pmset commands to block sleep only during agent activity, resumes sleep when tasks complete, and includes features like a status indicator and manual control via an optional MCP. The app is open-source under the MIT license and notarized.
- OpenAI says employees moving beyond chat to agents
OpenAI reports that employees are increasingly using AI agents like Codex over chatbots for multi-step tasks, with non-developers driving significant growth. External organizations and individual users also show rising adoption, though at lower rates. The shift correlates with higher token usage and potential financial benefits for OpenAI.
- Exclusive: Codex agents are inching into the mainstream
Codex, OpenAI's agentic coding platform, is gaining mainstream traction as usage shifts from chat-based AI to delegated work. A report by OpenAI, Columbia, Duke, and the University of Pennsylvania reveals 99.8% of OpenAI employees use Codex for most output tokens, with organizations and individuals showing rising adoption. Non-developers are the fastest-growing user group, and agents now handle tasks like managing calendars, files, and scripts.
- Codex logging bug may write TBs to local SSDs
A logging bug in Codex may cause it to write terabytes of data to local SSDs. The issue is discussed in a GitHub issue and a Hacker News thread with 18 points and 3 comments.
- Forget prompt engineering: 'Loop engineering' is all the rage now
Loop engineering is gaining prominence in AI as a replacement for prompt engineering, with figures like Claude Code creator Boris Cherny and OpenClaw creator Peter Steinberger advocating for automated systems that reduce manual prompting. Loops enable AI agents to perform recurring tasks autonomously, such as coding and repository maintenance, by structuring workflows with components like automations, worktrees, and sub-agents.
- Try One of macOS 27’s Best Features Right Now
Apple's upcoming macOS 27 release will allow users to build Shortcuts by typing commands. Users of Claude Code and Codex can access this feature immediately without waiting for the official release.
- Why right now might be the best opportunity to get hired at Silicon Valley's most coveted employers
Anthropic and OpenAI are emerging as the most sought-after employers in Silicon Valley, offering prestige, stability, and growth amid the AI boom. Applicants are flocking to these AI labs for opportunities to shape technological innovation and potentially benefit from anticipated IPOs, with both companies rapidly expanding their workforces.
- Nvidia Built Robots That Train Themselves Using AI Coding Agents
Nvidia developed a system called ENPIRE that allows coding agents like Codex and Claude Code to autonomously write training code for robots, test it on real hardware, and improve without human oversight.
- Research pulls back curtain on Claude
A research paper reveals that the orchestration layer, which handles non-AI operational tasks, is more critical than AI models themselves for real-world functionality. The study highlights that only 1.6% of Claude Code's codebase involves AI decision logic, with the rest dedicated to operational infrastructure.
- Good news, vibe coders: OpenAI says Codex is back to normal after experiencing 'elevated errors'
OpenAI's Codex AI coding tool experienced 'elevated errors' and 'model at capacity' issues, prompting user complaints. OpenAI confirmed the problem and reported full recovery by 9:48 a.m. Eastern Time. Users shared social media posts about the disruptions, and the company indicated ongoing efforts to stabilize the service.
- OpenAI is buying a cloud startup to supercharge its AI coding tool
OpenAI is acquiring Ona, a cloud startup, to enhance its AI coding tool Codex. Ona's secure cloud environments will enable Codex to handle longer-running tasks.
- OpenAI to acquire Ona to support its AI coding assistant, Codex
OpenAI is acquiring Ona to enhance its AI coding assistant, Codex. Ona's technology will enable Codex to handle longer-running tasks, according to OpenAI.
- Confessions of an AI lab rat
Axios CEO Jim VandeHei details his year-long experiment using AI tools like OpenAI's Codex and Anthropic's Claude across Axios operations and personal data. He highlights AI's superior accuracy and creative capabilities but notes its challenges in real-world application. VandeHei emphasizes AI's potential for new business opportunities over job losses and shares his use of AI for health diagnostics.
- Proliferate (YC S25) is hiring to building open source Codex
Proliferate, a YC S25 company, is hiring a founding engineer to build an open-source Codex. The job posting emphasizes contributions to an open-source project.
- OpenAI readies ‘superapp’ pivot ahead of planned IPO, FT reports
OpenAI is reorganizing ChatGPT into a 'superapp' to compete with rivals like Anthropic ahead of its planned IPO. The overhaul includes adding coding tools, image generation, and partner applications such as Canva and Booking.com, while prioritizing business customers who now account for 40% of revenue.
- Show HN: Lathe – Use LLMs to learn a new domain, not skip past it
Lathe is an LLM-powered tool that generates hands-on, source-backed tutorials for technical topics. It combines a Go CLI with LLM agent skills (Claude Code, Cursor, Codex) to create interactive learning experiences, encouraging users to type code manually and engage with exercises, side-notes, and citations. The tool aims to fill gaps where human-written tutorials are unavailable, such as niche domains like building a 3D slicer in Erlang.
- ChatGPT is no longer OpenAI's most important product. Here's why.
OpenAI and Anthropic are shifting focus from model quality to customer lock-in, prioritizing profit margins over performance. Products like Claude Code and Codex aim to create sticky, high-margin ecosystems, but enterprises like Walmart are developing tools to avoid vendor lock-in.
- Harness engineering: Leveraging Codex in an agent-first world
The article discusses harness engineering, focusing on the use of OpenAI's Codex in an agent-first world. It highlights the integration of Codex into engineering practices centered around autonomous agents.
- Supabase doubles valuation to $10B in 8 months
Supabase, an open-source project turned fast-growing company, has doubled its valuation to $10 billion in eight months, largely due to the benefits from AI tools like Claude, Codex, and other vibe-coding platforms.
- Ask HN: High school student – is learning programming still worthwhile?
A high school student questions whether learning programming remains worthwhile given the rise of AI-powered tools like Claude Code and Codex, which can automate coding tasks. The student acknowledges the influence of entrepreneurs such as Elon Musk and Jensen Huang but wonders if programming skills will stay relevant in the next decade as AI evolves.
- OpenAI's agent chained decade-old DoS attacks to crash web servers in seconds
OpenAI's Codex agent discovered a new denial-of-service (DoS) attack named HTTP/2 Bomb, which combines two decade-old techniques to crash vulnerable web servers in seconds. The exploit affects default HTTP/2 configurations on servers like nginx, Apache, and Microsoft IIS, with some vendors having released patches while others dispute the findings.
- Show HN: Boxes.dev: ditch localhost; run Claude Code and Codex in the cloud
Boxes.dev is a cloud-based agentic dev environment (ADE) developed by Nick and Drew, enabling users to run Claude Code and Codex agents on remote compute. It addresses localhost limitations like git worktree complexity and mobile coding challenges, offering a desktop and mobile app with features like isolated testing environments and Slack integration.
- MoonPay Brings Crypto Transactions to Claude and Codex With MoonAgents Desktop App
MoonPay has launched a desktop app that connects AI assistants like Claude and Codex to crypto wallets and blockchain services via a graphical interface. The app, called MoonAgents, aims to streamline crypto transactions for users of these AI tools.
- OpenAI launches new Codex tools for white-collar work
OpenAI has launched new Codex tools aimed at enterprise users to expand their use in white-collar work. The company also released an internal report highlighting Codex's applications in knowledge work.
- Exclusive: Office workers embrace OpenAI's Codex
OpenAI's Codex now has 4 million weekly active users, with knowledge workers comprising one-fifth of users and growing three times faster than developers. Key tasks include data analysis, research, and managing workplace artifacts, while users report both productivity gains and mental fatigue from managing AI workstreams.
- OpenAI frontier models and Codex are now available on AWS
OpenAI's frontier models and Codex are now available on Amazon Web Services (AWS). This collaboration enables developers to access advanced AI models through AWS's cloud platform.