Jack Dorsey now asks AI agents what to do instead of telling them

Jack Dorsey changed how he works with AI agents, and it reads like a glimpse of where agent workflows are going.

From commands to questions

His post on X, dated July 10, 2026, put it plainly: “i’ve shifted from telling agents what to do, to asking them what to do, and pulling the best thread.”

Instead of micromanaging one agent through every step of a task, Dorsey lets agents propose directions, explore competing paths, and then he picks the strongest idea. Less command line, more suggestion box, with a human still making the final call.

It is a small change in phrasing and a large change in posture. In the first model, the human supplies the plan and the agent supplies the labor. In Dorsey’s model, the agents supply candidate plans, and the human’s job compresses into selection and judgment.

The Block context

The shift fits the broader argument Dorsey has been pushing at Block. According to Fortune, he has argued that AI can absorb much of middle management, and that companies should run on a living “world model” of their operations. If agents can hold that model and generate options from it, asking them what to do stops sounding like a gimmick and starts sounding like delegation.

The open question

Supporters call this the smart way to use agents for brainstorming and discovery: fan out cheap explorations, keep the best one, discard the rest. Critics ask the obvious question in response: when agents set the direction, who is actually exercising judgment?

That tension is not going away. As agents get better at proposing plans, the temptation to let them frame the problem, not just solve it, will only grow. Dorsey’s answer, for now, is that a human still pulls the thread. How long that remains the default is the interesting part.

So which mode are you in: telling your AI what to do, or asking it?