Mythos is drawing attention for something most AI coding and task agents still can't do reliably: operate in a live environment outside a sandboxed container, then notify you when the job is finished.
Most AI agents run inside a sandbox - an isolated, contained environment that has no access to your actual file system, external APIs, or production systems. It keeps things safe but also limits what the agent can actually accomplish. Mythos can step outside that boundary, interacting with real systems, and then send you a notification when the task is complete. The practical result: you queue up a task, go to lunch, and come back to a finished result rather than a spinning progress bar.
The async notification piece matters more than it sounds. The biggest friction with current AI agents isn't their capability - it's babysitting them. Most require you to stay at your desk and monitor the run. A tool that handles real-world execution and reports back when done fits much closer to how actual work gets delegated.
Mythos appears aimed at developers and technical users who want agents that can touch real environments, not just demo-safe test beds. The sandbox-escape capability also raises the obvious questions about what guardrails are in place when the agent has real system access - something worth understanding before pointing it at anything critical.
No pricing or availability details were part of this showcase.