TL;DR
AI use is widespread, while separate McKinsey & Company and Deloitte surveys show more limited agent scaling and process preparation. Before a pilot, our recommendation is to check one workflow’s ownership, handoffs, data access, exceptions, human review and baseline.
- 1. McKinsey & Company's 2026 survey reports agent scaling at 40% of organizations above $1 billion in annual revenue and 22% of smaller organizations.
- 2. Deloitte's 2026 survey of US organizations at least piloting agents found 16% reporting prepared business processes and 5% highly prepared.
- 3. Gartner forecasts that over 40% of agentic AI projects will be canceled by the end of 2027; this is not an observed failure rate.
- 4. Our six checks are practical recommendations for a pilot, not a validated readiness score.
On this page
Nearly nine in ten respondents to McKinsey & Company's State of AI survey, published in August 2026, report regular AI use. Far fewer are scaling AI agents: 40% at organizations with more than $1 billion in revenue, and 22% at smaller ones, where the share stayed flat.
Even the companies already trying agents say their processes aren't ready. In a Deloitte survey of US organizations at least piloting agents, only 16% said their business processes were prepared for agentic AI, and just 5% said they were highly prepared.
What the Surveys Actually Measure
These numbers come from different studies asking different questions.
Use. McKinsey's 2026 survey of 1,719 participants in 97 countries found that nearly nine in ten report regular AI use in at least one business function. A year earlier, its 2025 survey put that figure at 88%.
Scaling agents. In McKinsey's survey, scaling means an agentic system is being scaled somewhere in the enterprise, not that agents run the whole company. In 2025, 23% reported that, and a further 39% were experimenting.
Process readiness. Deloitte surveyed 501 US senior managers and executives between April and June 2026, all at organizations at least piloting agentic AI, and published the results in August 2026.
Don't stack these into one funnel: the surveys cover different years, countries and companies. Together they show a direction. Using AI is common, scaling agents is not, and even the companies trying say their processes aren't ready.
Don't Expect an Agent to Fill the Gaps
Agents don't just answer questions. They act inside a workflow: pull data, decide at a fork, hand work to the next person or system. Those are the points an unready process leaves open.
Before a pilot, make ownership, handoffs and escalation paths explicit. Do not assume an agent will resolve an undocumented handoff or know when to escalate. We've made this case before: AI won't fix bad workflows, it makes them faster.
The cost of skipping that groundwork shows up later. Gartner forecasts that over 40% of agentic AI projects will be canceled by the end of 2027, and attributes that to costs, unclear business value or inadequate risk controls. That is a forecast, not a count of failed projects, but every one of those reasons is something you can check before you start.
A Readiness Check Before You Buy an Agent
This is our own checklist, not a validated score. Run one workflow through it before you sign anything.
- Map the inputs and handoffs. Write down what starts the workflow, every step, and where work passes from one person or system to another. If three people describe it three different ways, stop here.
- Name an owner. One person, not a team, accountable for the workflow and for what the agent does inside it.
- Test data access and permissions. Can the agent reach the data it needs, and only that? Check with real records, not a demo.
- List the exceptions. What happens when an input is missing, wrong or unusual? Every exception needs a path, and a person who receives it.
- Decide where a human reviews. Pick the points where a person approves, corrects or can stop the agent, before it goes live.
- Define the outcome you'll measure. Baseline the workflow today: time, cost, error rate. Without that, you can't tell whether the agent helped. Our AI automation ROI framework walks through it.
If a workflow fails any of these, fix that first. If it passes all six, it's a candidate for a small, measured pilot.
Our Take: Readiness Is the Work, Not the Obstacle
AI adoption is outpacing agent readiness, and companies are saying so themselves. The surveys don't prove that processes are the only reason agents stall. Cost, unclear value and weak risk controls matter too.
But process work is the part you control. Treat process design as a separate workstream when you buy software. The companies that scale agents won't be the ones that bought the most of them. They'll be the ones that knew exactly which workflow each agent was for, who owned it and how they'd know it worked.
We'd change our view if new data showed smaller firms scaling agents without that groundwork.
What We Don't Know
We don't know how many companies overall have processes ready for agents: Deloitte's figure covers US organizations at least piloting agents, not everyone. The surveys can't tell us how much of the scaling gap comes from process readiness rather than cost, skills or risk. We don't know whether Gartner's cancellation forecast will hold. And none of these studies splits process readiness by company size, so we can't say whether smaller firms are less ready or simply slower to invest.
This Quarter: One Workflow, Six Checks
Pick one workflow that matters and runs often. Put it through the six checks above. Fix what fails. Then pilot an agent on that workflow alone, with a named owner, a human review point and a baseline to measure against.
For the governance side, our AI governance checklist for operational leaders covers ownership, inventory and risk review.
FAQ
How many companies are scaling AI agents?
McKinsey & Company's 2026 State of AI survey reports 40% among respondents at organizations with annual revenue above $1 billion, compared with 22% at smaller organizations. Scaling somewhere in a business does not mean agents run the whole company.
Are business processes ready for AI agents?
In Deloitte's 2026 survey, 16% reported prepared business processes and 5% highly prepared. The sample covered 501 US senior managers and executives at organizations at least piloting agentic AI, so these figures do not describe all companies.
How should I assess a workflow before an agent pilot?
Our recommendation is to map handoffs, name an owner, test data access, list exceptions, set human review points and record a baseline. These checks guide preparation; they are not a validated score or a guarantee of success.