Codex Gains Early Traction as OpenAI Pushes Agent-Led Coding
After 200,000 first-day downloads, OpenAI CEO Sam Altman says building with AI agents left him briefly “a little useless,” highlighting a shift from writing code to supervising it.
Topics
News
- Codex Gains Early Traction as OpenAI Pushes Agent-Led Coding
- Supreme Court Raises Red Flags Over Meta, WhatsApp Data Practices
- Musk Merges SpaceX and xAI in Deal Valued at $1.25 Trillion
- Firefox Moves to Make AI Optional for Users
- OpenAI, Snowflake in $200 Million Deal to Embed AI in Enterprise Data
- Oracle to Raise Up to $50 Billion in 2026 to Scale Cloud Capacity
OpenAI’s new Codex coding app has generated early momentum, with the company saying it was downloaded by more than 200,000 users on its first day and gaining traction among developers.
The macOS desktop app is designed as a “command centre” that lets programmers manage multiple AI agents, run tasks in parallel and collaborate on long-running coding projects.
Sam Altman said on X that more than 200,000 people downloaded the Codex app on its very first day, adding that early users “seem to love it.” But alongside the celebratory “CODEX FTW,” Altman struck a noticeably introspective note about what tools like Codex could mean for human creativity and usefulness.
OpenAI Chief Executive Officer Sam Altman marked the launch with a candid post on X, saying early users “seem to love it,” but reflecting on his own experience building and extending an app with Codex.
“At least a couple of them were better than I was thinking of,” he wrote. The experience, he admitted, left him feeling “a little useless,” and even “sad.”
While expressing confidence that society will find “much better and more interesting ways to spend our time,” Altman said he feels “nostalgic for the present”, a rare moment of vulnerability from a tech leader usually focused on the future.
The comments came days after OpenAI officially introduced the Codex app for macOS on February 2.
Positioned as a “command center for agents,” the desktop app is designed to help developers manage multiple AI agents simultaneously, run tasks in parallel, and collaborate with agents on long-running projects that can stretch over days or even weeks.
OpenAI says Codex reflects a broader shift in software development. Since Codex launched in April 2025, models have become capable of handling complex, end-to-end tasks, changing the challenge from what AI agents can do to how humans supervise and direct them at scale.
The app allows developers to run agents in separate threads organized by project, and use built-in worktrees so multiple agents can work on the same repository without conflicts. Each agent operates in an isolated environment, letting developers explore different approaches without disturbing their local codebase.
Codex is also expanding beyond code generation through “skills”, bundled instructions and workflows that allow agents to gather information, solve problems, write, and interact with other tools.
OpenAI showcased this by having Codex independently build a complete racing game, using image generation and web development skills, consuming more than 7 million tokens from a single prompt.
OpenAI has positioned the app as part of a wider effort to make agentic tooling more accessible, temporarily including it with ChatGPT Free and Go plans and increasing usage limits across premium subscriptions. Early usage data also shows that overall adoption of Codex has expanded rapidly, with more than a million developers reported to have used the tool in recent months.
Altman’s candid post also sparked criticism from sections of OpenAI’s user community, particularly around the recent shutdown of GPT-4o. One user on X pushed back sharply, writing, “You played with an app for a weekend and felt ‘sad.’ We built lives, memories, and emotional anchors with 4o for years, and you deleted them for profit.” The user called Altman’s “off-script” moment “a slap in the face,” adding, “Integrity > Ego.”
Another user questioned the timing and intent behind both the Codex launch and Altman’s remarks, asking, “What exactly is the intention here, was this meant to distract us?” While acknowledging that launching new products is normal, the user said doing so while people are still seeking clarity on the GPT-4o shutdown “just feels tone deaf.” The post ended with a direct appeal: “Respond to the real questions. Please reconsider 4o shutdown.”
The launch comes amid intensifying competition in AI-powered coding tools, with rivals such as Anthropic’s Claude Code already gaining significant enterprise traction.
While these tools can enhance productivity, they have not yet replaced traditional development environments, and adoption patterns vary by workflow and team needs, analysts said.
