Deloitte asked more than 500 US business and IT leaders, all directly involved in their company's AI agent plans, how ready they actually are for autonomous AI agents. The answer: not very. Only 1 in 5 said their organization is prepared to redesign how work gets done for agents that can act on their own.
That gap is striking because expectations are sky high. Nearly three in four leaders expect close to half of their business processes to be rebuilt around AI agents within four years. Most also expect those agents to run with little human input, coordinating across departments and finishing multi step jobs on their own.
Here is the part that should worry any operator watching from the outside: even the companies furthest along do not trust their own readiness. Among the group Deloitte calls scaled adopters, the businesses already running AI agents widely, fewer than half believe their processes are actually set up for it. If the leaders are not confident, everyone else is further behind.
Deloitte's warning is that many companies are taking a shortcut called layering, which means bolting an AI agent onto an existing process without changing the process itself. That can produce a quick, visible win. But it does not hold up once the agent needs to make real decisions, and it leaves the harder work of redesigning jobs and oversight for later, or never.
This pattern shows up everywhere else researchers have looked recently. KPMG found companies pulling back on agent projects once the costs ran past the payoff, and in one case an audit found a company running ten times more AI agents than its own leadership realized. Nobody was tracking what they had built, and the bill showed up before anyone noticed.
A Federal Reserve Bank of St. Louis study of company earnings calls found executives are talking about AI's effect on productivity constantly and almost entirely in glowing terms. But actual company wide productivity numbers have not moved much yet, which means the payoff executives describe is still mostly a promise rather than a result already booked.
Inside HR departments specifically, a Culture Amp survey found confidence in AI actually dropped this year compared to last year. Most HR teams are still using AI as a smart assistant rather than letting it operate with real independence. The teams that did make that jump saw process improvements two to three times better than everyone else, which suggests the gap is not about the technology working, it is about companies being unwilling or unprepared to hand over real authority.
The practical takeaway is simple. Know exactly what AI agents are already running inside your company, who is accountable for their decisions, and whether your people have real time and training to work alongside them. The businesses that redesign the actual work, not just add a tool on top of it, are the ones that will pull ahead, while everyone else risks paying for agents that never deliver the returns leadership promised the board.