OpenAI’s Codex has quietly crossed a milestone that puts the “AI agents” narrative firmly into the present tense. The platform now counts more than 3 million weekly active users, a figure that grew more than fivefold since January 2026.

From autocomplete to autonomous worker

An April 16, 2026 update introduced multi-agent parallel execution, computer control capabilities, and the ability to delegate tasks across the entire software development lifecycle. Codex can now run multiple AI agents simultaneously, each handling different parts of a project, clicking through interfaces, and handing work off to one another.

Tasks in its sandbox environments typically complete within minutes, with more complex jobs wrapping up in about 30 minutes. The platform supports both cloud-based and local execution through CLI and IDE integrations.

On June 2, 2026, OpenAI released a report highlighting Codex’s expansion beyond software engineering into automating routine knowledge work across various professions.

Enterprise heavyweights are already in

Goldman Sachs, DoorDash, Thermo Fisher, State Farm, and LY Corporation are all actively expanding their use of the platform in 2026.

NVIDIA has noted outstanding performance improvements when pairing its infrastructure with GPT-5.5, the model powering Codex’s latest agentic workflows. OpenAI has emphasized that these newer agentic capabilities represent a fundamental shift toward AI-driven operational efficiency at scale.

What this means for crypto and decentralized compute

Codex has no direct connection to cryptocurrency protocols, tokens, or blockchain projects. But the explosion of AI agent workloads creates massive demand for computational infrastructure. Every Codex task, every parallel agent execution, every enterprise deployment consumes GPU cycles at scale.

Projects like Render Network, Akash, and io.net have positioned themselves as alternatives to centralized cloud providers, aggregating idle GPU capacity into marketplaces for AI workloads. Most corporate IT departments aren’t going to route Goldman Sachs’ AI workloads through a permissionless GPU marketplace anytime soon. But at the margins, for less sensitive workloads, for smaller companies priced out of hyperscaler contracts, the opportunity is real.

NVIDIA’s close partnership with OpenAI suggests the centralized compute stack is well-positioned to capture most of the value. Decentralized alternatives need to prove they can compete on performance, not just price.