Ask a leadership team whether the company uses AI and the answer is almost always yes. Ask which process now runs differently because of it, who owns that process, and which number moved, and the room goes quiet. That gap, between using AI and putting it to work, is where most of the 2026 data points.
Using AI and putting it to work are two different things
Using AI means a person asks a question and gets a better answer: a summary of a long thread, a first draft of an email, a formula in a spreadsheet. It is useful, and it is where almost every company starts. Putting AI to work means a business process runs differently: a supplier invoice is matched and routed without anyone retyping it, an IT ticket is resolved before it reaches the queue, a quality report is assembled from the line data overnight. The first saves minutes for individuals. The second changes a number someone is accountable for.
The 2026 data shows most companies still sitting on the first side. In Gallup's survey of 22,573 US employees, 52% use AI in their role, but only 15% use it daily.1 In Microsoft's 2026 Work Trend Index, 86% of AI users treat what AI produces as a starting point rather than a finished piece of work, and only 14% of teams outside the most advanced group have AI-assisted workflows documented and repeatable across the organization.2 McKinsey finds 37% of companies attributing any EBIT impact to AI, unchanged from last year.4
None of this means the tools are failing. It means that a tool on everyone's desktop is not the same as a process that runs on it.
The four levels of AI use
We use four levels to describe where a company stands. They are not a maturity model to be completed in order over years; they are a way to name the current state honestly, so the next step is obvious.
Asking
People use ChatGPT, Copilot or similar on their own to ask, summarize and draft. No process runs differently.
Signal: nobody can say what the company spends on AI or who owns it.
Piloting
Copilot or another tool is switched on for a group. People like it. The business hasn't changed yet.
Signal: success is measured in licenses assigned and active users.
Building
At least one process runs on AI in production: an agent or workflow with a business owner and a number it moves.
Signal: someone outside IT asks for the next one.
In production
Several processes run on governed agents, with security, an inventory and a roadmap. Value is measured against cost.
Signal: stopping an agent that doesn't pay is a routine decision.
Where companies are in 2026
No single survey measures the four levels, so read the figures below as a pattern rather than a funnel. They come from different studies with different definitions.
Sources: Gallup Q2 2026 (n=22,573 US employees), rows 1 and 4; KPMG Global AI Pulse Q3 2026 (n=2,131), row 2; Gartner functional leaders survey (n=1,303, Sep 2026), row 3; Gartner, Apr 2026, row 5. Different surveys and definitions; read as a pattern, not a funnel.
Level 1 and 2: where most companies are
Half of the 20,000 AI users in Microsoft's study fall into what it calls the "emergent" zone: using AI, but without the team practices that turn it into new ways of working. Only 19% are in the frontier group.2 Among manufacturers, Rockwell Automation's 2026 survey of 1,560 companies found 18% still in pilot mode with smart manufacturing technologies, and only 43% of the data they collect being used effectively.10 Pilots look good on their own terms: 97% of middle-market companies rate them a success, yet only 17% are pursuing change across the enterprise.9
Level 3: agents are arriving faster than the rules for them
Agents, AI that carries out steps in a process rather than answering a question, moved from demo to daily use this year. 34% of organizations in KPMG's September survey report significant employee adoption of AI agents, up from 25% in the first quarter.3 Active agents in Microsoft 365 grew 15 times in a year.2 Size still matters: 40% of large organizations are scaling agents, against 22% of smaller ones.4 In manufacturing, 56% of middle-market companies in RSM's 2026 survey already use agentic AI in some form.9
Level 4: few are there, and governance is the reason
Only 22% of organizations have scaled AI across more than one business unit.5 Only 13% believe they have the right governance for AI agents, and Gartner expects the average Fortune 500 company to have more than 150,000 agents in use by 2028.6 IBM's CIOs and CTOs averaged 54 agent incidents last year that needed human correction, and only 11% say they are fully ready for the agent deployment they expect.8
What changed in the tools this year
In 2026 the software moved from answering to doing, and much of it now sits inside the Microsoft 365 plan companies already pay for. For a business leader, four changes matter:
- Copilot Chat does more without an extra license. Since the July 2026 packaging update, the Copilot Chat included in Microsoft 365 plans is aware of the user's inbox and calendar and gives access to Word, Excel and PowerPoint agents.11
- Building an agent no longer needs a development project. Users with a Microsoft 365 Copilot license can build and use internal agents on company data with Copilot Studio at no additional cost. Copilot Business, for companies up to 300 users, lists at $21 per user per month.12
- Agents that act are billed by use. Advanced experiences such as Copilot Cowork and Copilot Autopilot are charged through Copilot Credits, with spending controls in the Microsoft 365 admin center. This is a new cost line that needs an owner.12
- Governance became a product. Agent 365, generally available since May 1 at $15 per user, gives IT one place to observe, govern and secure agents; Microsoft 365 E7 bundles it with E5 and Copilot at $99 per user.13
Manufacturers are already using these pieces in production. Microsoft reports that AGCO has several hundred employee-built agents in production on Copilot Studio, and that in some quality workflows reviews that used to take weeks now take about an hour. These are customer-reported results published by Microsoft.14
The technology reached Level 3 and 4 faster than most organizations did. The constraint in 2026 is not what the software can do; it is ownership, data and governance.
How to move up one level
- From Asking to Piloting: pick one process, not ten use cases. Work the team does every week, with a visible cost today: hours spent, tickets waiting, a task that depends on one person.
- From Piloting to Building: name a business owner and the number that should move before you start. Measure the process, not the logins. Redesign the steps around the AI rather than adding AI to the old steps.
- From Building to In production: keep an inventory of every agent, give each one an identity and an owner, decide what data it may see, monitor what it does and retire the ones that don't pay. These are the controls Gartner recommends for agent sprawl.6
Manufacturing AI Executive Breakfast: From AI Potential to Business Outcomes
Microsoft and Bravent host an executive discussion and a short workshop to identify and prioritize the highest-impact AI opportunities in your operation. Not a product presentation.
Wednesday, December 2, 2026 · 8:30–11:00 AM ET · Microsoft office, 830 Brickell Plaza, Miami
Want to know what this looks like in your company?
Tell us who you are and what you'd like: more information about the Switch It On program, or a meeting with the Bravent team to look at your Microsoft 365 tenant together. Someone from Bravent will get back to you.
The AI you already pay for: what's inside your Microsoft 365 →
The twelve blocks in a typical plan, what changed on July 1, 2026, and which ones to switch on first.