The state of AI in 2026: everyone has adopted it, few have scaled it

Everybody bought the shovel. That's the short version of the last two years of AI at work: a gold rush, tools that got cheap and frighteningly good almost overnight, every company grabbing one. The digging, the part where it turns into money, has barely started. Here are the questions people keep putting to me about where AI stands right now.

Almost all of them. McKinsey's State of AI research has 88% of organisations using AI somewhere in the business, and 79% using generative AI specifically. What gets me is the speed. On McKinsey's own earlier figures, that 79% was 65% at the start of 2024, and about a third the year before. We skipped the cautious pilot phase entirely and went straight to "why haven't we got this everywhere?"

Mostly, not yet. Far fewer companies have scaled AI across the whole business than have adopted it, and a much smaller group can point to a clear lift in revenue.

Everyone owns a shovel; few have struck anything. I've sat in rooms where a leadership team was proud they "have AI now", asked what it had changed, and watched the pause land. That pause is the whole problem.

Because the tool gets bolted onto how the company already works, instead of changing how it works. A chatbot on the site. A copilot in three people's inboxes. Useful on their own, but they don't talk to each other, and a drawer full of clever gadgets isn't a machine. A system is.

You can usually spot the difference on an ordinary Tuesday. In the gadget version, someone copies an enquiry from the inbox into the CRM, asks the chatbot to draft a reply, pastes it back and forgets to set a reminder. In the system version, the enquiry lands in the CRM, a reply goes out in a minute, the right person gets an alert and a follow-up is already scheduled. Same tools. Very different week. Our AI email sales agent case study is the system version, running a sales inbox end to end.

It's the same wall people hit with agentic AI and automation, only there the falls cost more.

They treat it as a decision about how the business runs, not a line on the software invoice. In practice:

1. Pick one or two places where it clearly earns its keep, usually where people repeat the same steps every week or customers wait on you.
2. Rebuild that part of the process around it, rather than adding AI as an extra step on top.
3. Measure before and after, then go back and check the number moved. This is the step everyone skips.

Owning a tool versus running a system: that's the whole game, and it's how we approach AI consultation and the work we do as an AI automation agency in Dubai. It's also why smaller firms are catching the big ones so fast, which is the story of how SMEs are actually using AI to grow.

AI isn't overhyped, and it isn't a bubble. Owning the shovel was just never the hard part. McKinsey's adoption numbers aren't a story about technology. They're a story about follow-through: the businesses getting value picked a specific job, built a system around it and checked it worked.

If your business has the tools and is still waiting for the results, we'll map where AI would change a number you care about, and what to build first.

Source: McKinsey, The State of AI.