ChatGPT Work Kills the Cracked-Laptop Workaround for AI Agents

2. ChatGPT Work Kills the Cracked-Laptop Workaround for AI Agents

Greg Brockman posted on July 19, 2026 that ChatGPT Work runs its agent sessions fully in the cloud, meaning tasks execute and persist whether the user's laptop is open, closed, or absent entirely. Access works from mobile. No local process required. Brockman framed this as a quiet but long-overdue fix, noting that leaving a laptop "cracked open" has been the de facto workaround for getting agent workflows to run continuously.

The competitive implication is direct. Products like Anthropic's Claude with computer use, Google's Project Mariner, and a wave of third-party agent frameworks have all shared the same structural constraint: persistent execution depended on a live local session or a developer-managed cloud wrapper the end user had to set up themselves. OpenAI is now absorbing that infrastructure layer into the product, which shifts the category from power-user tool to something closer to a background service. That changes who can realistically use agents day-to-day, and it raises the barrier for competitors whose agent products still require an open browser tab or a running desktop client to stay alive.

The broader pattern here is productization pressure compressing the gap between research capability and daily usability. Cloud-native execution is not a new idea in software, but AI agent products have been slow to close this gap because session state, tool calls, and long-running tasks are harder to manage server-side than a stateless chat response. OpenAI shipping this in ChatGPT Work signals the infrastructure problem is solved enough to ship. Watch whether Anthropic or Google announce equivalent always-on, cloud-native agent execution in their consumer or enterprise tiers before the end of Q3 2026.

Source: Greg Brockman on X