$ today_in_10_seconds
A new Riverbed global survey puts numbers on the central tension in enterprise agents. 90% of IT orgs plan to pursue autonomous IT, yet 77% still require a human to approve what the agents do. The demand is there, and trust is now the constraint.
$ lead_story
90% Plan for Autonomous IT, but 77% Still Demand a Human Sign-Off
Riverbed's new global survey finds agentic AI is reshaping IT operations. Nine in ten organizations say they plan to pursue autonomous IT, and more than three-quarters say a human must still approve agent actions. Those two figures are less contradictory than they look. Together they describe what enterprises are actually buying right now: agents that diagnose, plan, and propose, while humans keep the final click. For builders, this means 'fully autonomous' is a roadmap slide, not the product you sell. The product you sell is a fast, legible, auditable approval loop.
// why_it_matters: In enterprise IT, the approval workflow is the agent product.
$ bigger_picture
The trend behind the headlines
The gap between 90% ambition and 77% gatekeeping isn't hesitation. It's the price of earning trust. Autonomy in production gets granted one action class at a time, starting with low blast-radius fixes and expanding as track records build up. The teams that win will treat human approval as a data source: every approve, reject, and edit is a labeled example of what the agent should have done. Instrument that loop well, and the 77% shrinks on its own schedule.
LangGraph
LangGraph is an open-source framework for building stateful agents. Its interrupt mechanism pauses a run mid-graph, surfaces the proposed action to a human, and resumes from a persisted checkpoint once someone approves or edits it. If your customers are in the 77%, this is a clean way to build the approval gate into the agent's control flow rather than adding it on afterward.
$ subscribe --free
Get the next edition in your inbox
// DAIly Agent covers agentic AI three times a week
// free, no spam, exit anytime