AI agents are the current obsession of corporate technology budgets. Unlike the chatbots most companies rolled out over the past few years, which just answer questions or draft text, agents are meant to plan and carry out multi-step work on their own, like handling a customer complaint from start to finish without a human writing out every step.
Two new surveys, one from Deloitte and one from Accenture, show that the gap between what companies hope for and what they are actually getting is still wide.
Deloitte surveyed tech leaders across five industries and found that most expect it will take three to four more years before half of their business processes are rebuilt around agents working independently and together. That timeline matters because right now, most companies are taking a shortcut: they are dropping agents into their current workflows instead of redesigning those workflows around what the technology can actually do. That shortcut gets something running quickly, but it does not deliver the bigger payoff that comes from rethinking how the work gets done in the first place.
The readiness numbers back this up. Only 5 out of 100 companies surveyed say their business processes are highly prepared for agents, and just 15 out of 100 have scaled up agents that coordinate across different teams or departments. Even companies that are further along still struggle: fewer than half of the most advanced adopters believe their own processes are ready for this kind of technology.
Accenture's separate survey of thousands of executives found a similar pattern from a different angle. Most companies report that agents are helping employees get more done, with more than two-thirds of senior leaders saying the effect on productivity has been bigger than they expected. But when it comes to results big enough to report to the board, the picture is worse. Only 23 out of 100 companies said they had that kind of proof this year, down from 32 out of 100 just months earlier. Executives told Accenture the drop is happening because companies are spreading agents thin across too many small experiments instead of pushing a few clear use cases all the way through to a measurable outcome.
There is also a trust problem underneath all of this. Older automated software follows the same fixed steps every time, so a company can check its work once and trust it to repeat. Agents do not work that way. They figure out their own path to a result, which means someone still has to check whether that result is actually right, every time. That checking work is new, and most companies have not built the habits or the oversight needed to do it well.
For a business leader watching this from outside the tech industry, the lesson is not to slow down on agents, it is to stop expecting quick, board-ready wins from scattering them everywhere. The companies pulling ahead are the ones picking a small number of processes, rebuilding those processes properly around what agents can do, and proving the value there before expanding. Everyone else is likely to spend heavily over the next few years with little to show a board beyond productivity anecdotes.