Quick facts
- Nearly 9 in 10 organizations use AI in at least one business function (McKinsey, 2026).
- 40% of large companies are scaling AI agents. Among smaller companies it is 22%, the same as last year (McKinsey, 2026).
- Only 37% say AI has added to company profit, unchanged from 2025 (McKinsey, 2026).
- More than 40% of agentic AI projects are forecast to be cancelled by the end of 2027 (Gartner, 2025).
- Only about 130 of the thousands of vendors selling “agentic AI” offer the real thing (Gartner, 2025).
- 82% of organizations use AI agents, but only 44% have policies to secure them (SailPoint, 2025).
- 66% of companies that adopted AI agents report higher productivity (PwC, 2025).
Most AI agent projects stall because they start with the technology and not with a business problem. The pilot looks good in a demo. Then it meets messy data, old processes, security worries and rising costs, and nobody can show what it earned.
The numbers back this up. Gartner expects more than 40% of agentic AI projects to be cancelled by the end of 2027. McKinsey’s 2026 survey found that almost every company uses AI, yet only 37% see it in their profits.
The causes are well known, and each one can be fixed. This guide explains what an AI agent is, why projects get stuck, and what the companies that scale do differently.
What is an AI agent?
An AI agent is software that can plan and complete a task on its own to reach a goal you set. You tell it what you want done. It works out the steps, uses your tools and checks its own work.
AI agent vs chatbot
A chatbot answers. An agent acts. That one difference explains both the promise and the risk.
| Chatbot | AI agent | |
|---|---|---|
| What it does | Replies to a question | Completes a task from start to finish |
| Example | Tells a customer the refund policy | Checks the order, approves the refund and updates the record |
| Access it needs | Your help articles | Your systems and data |
| What can go wrong | A wrong answer | A wrong action |
Where agents are used today
Agents work best on tasks that are frequent, structured and easy to check. McKinsey found companies most often scale them in IT, knowledge management and software engineering. In PwC’s survey of US executives, the top areas were customer service (57%), sales and marketing (54%), and IT and cybersecurity (53%).
How many companies have scaled AI agents?
Far fewer than the headlines suggest. Using AI is common. Running agents across the business is still rare, and it is mostly large companies doing it.
| Where companies stand in 2026 | Share |
|---|---|
| Use AI in at least one business function | Nearly 9 in 10 |
| Scaling AI across the whole company | 44% |
| Large companies scaling AI agents | 40% |
| Smaller companies scaling AI agents | 22% |
| See AI adding to company profit | 37% |
| Count as AI high performers | About 6% |
Source: McKinsey Global Survey on the state of AI, 1,719 respondents in 97 countries, May to June 2026. “Large” means more than $1 billion in annual revenue.
The gap between use and results
People feel the benefit long before the company does. Eight in ten respondents told McKinsey that AI made them more productive. Yet the share of companies reporting a profit impact has not moved in a year.
PwC saw the same pattern. Most executives said their company was adopting agents. But 68% said half or fewer of their employees work with an agent on a normal day.
Why do most AI agent projects stall?
Five causes come up again and again in the research. Most stalled projects have more than one.
1. There is no clear problem to solve
Many projects begin because agents are exciting, and the business case comes later. Gartner describes most current projects as early experiments driven by hype and often misapplied. It adds that many tasks sold as “agentic” do not need an agent at all.
Vendors make this worse. Gartner calls it “agent washing”: old chatbots and automation tools renamed as agents. By its estimate, only about 130 of the thousands of vendors in this market are real.
2. The agent is added to an old process
Putting an agent on top of a broken process gives you a faster broken process. McKinsey found that nearly three-quarters of its AI high performers redesigned their workflows around AI. Only one-quarter of other companies did.
3. The data is not ready
An agent can only act on what it can reach. IBM research, cited in Hostinger’s 2026 statistics roundup, found that AI models can access just 31% of a typical organization’s own data and use only 13%. Records sit in separate systems, in different formats, with gaps.
4. Security and trust are an afterthought
An agent that can act can also act wrongly. In a SailPoint survey of 353 IT professionals, 80% said their agents had done something unintended. For 39%, that meant reaching systems the agent should not have touched.
Leaders notice. In PwC’s survey, 38% trusted agents with data analysis. Only 20% trusted them with financial transactions.
5. Costs grow before value shows
Pilots are cheap. Production is not. Gartner names rising costs as a main reason projects get cancelled. One in five respondents told McKinsey that AI running costs already limit how much they use it.
How to get an AI agent past the pilot stage
Companies that scale agents tend to do four things early.
Start with one workflow that hurts
Pick a task that happens often, follows clear rules and is easy to check. Customer support is a common first choice. Hostinger reports that its own support agent resolved 75% of about 750,000 monthly conversations without a human by August 2025, up from 50% at the start of that year.
Decide how you will measure success
Choose one number before you build anything. It could be tickets resolved, hours saved or cost per order. McKinsey’s high performers are twice as likely as others to have a defined way of measuring AI’s impact.
Redesign the process, then add the agent
Ask how the work should flow if an agent handles the routine steps. Remove steps that only exist because people did them by hand. Gartner says rebuilding a workflow from the ground up is often the better path, because fitting agents into old systems is complex and costly.
Give the agent clear limits
Treat an agent like a new employee. Give it access only to what the job needs. Keep a record of what it does. Require a person to approve anything involving money or sensitive customer data. SailPoint’s advice is to govern agents as strictly as human users.
When is an AI agent the wrong choice?
Sometimes the best decision is not to use an agent. Gartner offers a simple rule: use agents when decisions are needed, automation for routine workflows, and assistants for simple lookups.
In practice, an agent is probably the wrong tool when:
- The steps never change. Ordinary automation is cheaper and more predictable.
- People only need to find information. A good search tool or assistant will do.
- A mistake would be expensive and hard to undo. Keep a person in charge.
- The task is rare. The setup effort will outweigh the time saved.
The bottom line
AI agents can deliver real results. Most projects stall because of what surrounds the AI: the goal, the process, the data and the controls. Pick one real problem, fix the process and the data around it, set limits, and measure the result. Then expand from what you have proved.
Sources
- Gartner: Over 40% of Agentic AI Projects Will Be Canceled by End of 2027 (25 June 2025)
- McKinsey: The state of AI in 2026: On the road to ROI (25 August 2026)
- PwC: AI Agent Survey (308 US executives, April 2025)
- SailPoint: AI agent adoption research (28 May 2025)
- Hostinger: Agentic AI statistics 2026 (8 July 2026), used for the IBM data figures and Hostinger’s own support results
How many AI agent projects are expected to fail?
Gartner forecasts that more than 40% of agentic AI projects will be cancelled by the end of 2027. The main reasons are rising costs, unclear business value and weak risk controls. The forecast was published in June 2025.
What is agent washing?
Agent washing is when a vendor renames an existing product, such as a chatbot or an automation tool, and sells it as an AI agent. The product has no real ability to plan or act on its own. Gartner estimates only about 130 vendors offer true agentic AI.
Are AI agents safe to use in a business?
They can be, if you control what they can access. SailPoint found that 82% of organizations use agents but only 44% have policies to secure them. Limit each agent’s access, log its actions and keep people involved in high-risk decisions.
Should a small business use AI agents?
Yes, but start small. McKinsey found that scaling agents is still led by large companies, with smaller ones holding steady at 22%. A small business will usually get more from one well-chosen agent, for example in customer support, than from several experiments.
Do AI agents deliver a return?
They can. Among companies that adopted agents, PwC found 66% saw higher productivity and 57% saw cost savings. Company-wide profit gains are less common, which is why choosing the right task and measuring it matters.