Korva is a scroll of the internet’s best ideas.
Intelligence is part of making agents useful, including how long they can stay on a task. A lot of the rest is tactical. An agent working locally is convenient because it can see all the data on a machine, but an agent in the cloud takes a ton of cloud infrastructure, reliability work, and connections to the third-party systems where the data lives.
Done right, the payoff changes how a day goes. Start a task from the mobile app in Work mode, ride a long subway with no service, and come out to find the thing done. Nothing runs on the laptop. The agent keeps working in the cloud until it finishes, and then the person can chat with it. That works on every surface, not just mobile.
Why do agents need access to company tools?
A language model on its own sees only the text placed in front of it. To act on a company's real work, it needs connectors into the systems where documents, messages, and records live. A shared standard means one connector can serve many different assistants. Anthropic released the open Model Context Protocol in November 2024 as a standard for those connections, and OpenAI announced support for it in March 2025.
How does a cloud agent keep working offline?
A cloud agent runs on remote servers, so the phone only sends the request and later receives the result. Long jobs save their progress to storage as they go, which lets them continue past a dropped connection or a restarted machine. This design is called durable execution. Workflow systems such as Temporal and AWS Step Functions are built on the same idea.
More from Lenny's Podcast: Product | Growth | Career
