Autonomous agent (dots, Gemini Spark, Autopilot)

Agents & orchestration

An autonomous agent is an AI agent that works continuously, in the background, without waiting for an instruction at every step. It usually has its own cloud computer, a browser, connected applications and its own access rights, and it remembers what it has done. Each provider gives it a name: “dots” at OpenAI, introduced on 29 September 2026, Gemini Spark at Google, Copilot Autopilot at Microsoft, Grok Bot at xAI and Muse at Meta. Anthropic, for its part, has merged its Cowork agent and chat into a single Claude. At launch, dots are not open to Pro subscribers in Europe.

In practice at Gensai

Gensai sets the level of autonomy of its agents with each client: a list of allowed actions, access limited to what is needed, human validation on sensitive steps and a log of every operation.

When should a task be handed to an autonomous agent?

For long, repetitive tasks that are easy to check: monitoring, preparing files, following up requests, keeping several tools in sync. When a mistake is costly or goes unnoticed, a supervised agent that proposes and waits for validation remains the better choice.

Common mistake

Giving the agent the same rights as an employee. An autonomous agent should have its own access, reduced to the strict minimum, and leave a trace of every action. OpenAI acknowledges that its dots can make mistakes and recommends checking any work with significant consequences.