Is your business really using AI, or just asking it questions?

Ask most executives how their AI rollout is going and you'll hear something like: "Really well, actually. The team loves it."

And they're probably right. People are saving time, writing faster, getting answers they'd have spent an hour digging for.

The numbers back it up: 68% of AI value in US businesses currently comes from assistants and copilots. ChatGPT, Copilot, Gemini — the tools everyone's opened a browser tab for.

These are wonderful gains. The question is whether they're the ceiling or the floor.

Time saving isn’t the whole picture

Companies running mainly on GPT-style tools tend to measure AI value in hours saved per employee, per week. And that number looks good — typically two to ten hours per person, which adds up fast across a team.

But "hours saved per person" only measures one thing: how much faster an individual can do a task they were already doing.

It says nothing about the tasks nobody's doing at all — the invoice that sat in a queue, the exception that didn't get flagged, the approval that waited three days for someone to remember to chase it.

That's where most businesses are still losing time. And a faster human with a better writing tool doesn't fix any of it.

The difference isn't speed, it's who starts the work, and where it stops. A copilot waits to be asked. An agentic app just gets on with it. And that's where most of the untapped value actually lives.

The habits that form without anyone noticing

But there’s another issue that sneaks up on businesses.

The longer you run on copilot tools, the more everything around them gets built to match.

Processes assume a human is checking each step. Approval chains assume someone is always in the middle.

The whole organisation optimises around "what can I ask the AI?" rather than "what should the AI just be handling?"

By the time most businesses are ready to ask the second question, they've already built a lot of infrastructure around the first. Unpicking that later is a bigger project than it needed to be.

GPT-only companies aren’t just missing the agentic wave. They’re building organisations that will be difficult to change when they’re ready to catch it.

That said, slower is often smarter

None of this is an argument for rushing into something more autonomous. Most businesses that have moved fast on this front have found it harder than expected.

AI systems running unsupervised on top of critical processes without the right guardrails and checks can cause real damage. Getting the wrong answer at scale tends to do more harm than a slow human process. Sequence matters more than speed.

Mapping how decisions actually get made, getting process knowledge out of people's heads and into systems, understanding where the real bottlenecks are, is worth doing now, while copilots are still delivering.

The groundwork that makes the next step possible comes down to one thing: structure.

Right now, most businesses have their AI tools running in isolation — an automation here, a chatbot there, a copilot bolted onto the CRM. Each one is doing its own thing. None of them talking to each other.

That's fine when AI is just helping individuals. It becomes a problem when you want AI to run something end-to-end at the organisational level.

What actually makes that possible is having all your apps, data and processes running on the same underlying logic — one shared system where everything is connected and every action is traceable.

That's what DecidrOS does. Instead of a patchwork of tools, you get a single operating layer where marketing, sales, finance and ops all run on the same structure.

A new app plugs in and immediately has access to the same data and the same rules as everything else. Nothing operates in a silo. Nothing conflicts.

That's the foundation that lets AI do more than assist, it's what lets it actually run things. And it's the bit most businesses skip, which is exactly why the next step ends up harder than it needs to be.

What the businesses pulling ahead have in common

The 26% our research calls Trailblazers didn't get there by moving the fastest. More than three quarters have a clear strategy and roadmap. Nearly a quarter already have AI running across multiple parts of the business.

They started with copilots like everyone else. The difference is they also started asking the second question early — and built the infrastructure to answer it.

The 33% who are Tinkerers — trying things across specific workflows without a formal plan — are well placed. The experimentation has built real intuition. The next move is turning that into something the whole business can run on.

Download the Decidr US AI Readiness Index 2026 to see where your business stands — and what the next phase looks like.