Three years ago, a tool that could draft a decent email in seconds felt like magic. Today it's the baseline. Nobody's impressed anymore that software can write; they expect it.

The development happened fast, and it quietly moved the goalposts for what "good software" even means.

The change: Generative AI made every tool capable of sounding intelligent, so sounding intelligent stopped being the thing that mattered.

The real question now: not whether a tool has AI, but whether that AI knows enough about your business to be useful.

Why it matters: the businesses getting real value aren't using the most AI. They're using it where it already has the context to act on and not fabulate about it.

Everyone has access to the same model. Almost nobody has the same context

The interesting part of the last three years was the incredible speed at which models got smarter and – more surprisingly – how fast that intelligence became a commodity. Every vendor in every category now has some version of a generative AI feature. Impressive output stopped being a differentiator the moment it became table stakes.

What didn't become a commodity is context: the history, patterns, and detail specific to one business and its customers. A generative AI tool with no memory of your last ten conversations with a customer can write beautifully and still be useless in the moment that matters. One that already knows what was promised, what broke, and what's overdue doesn't need to sound as clever; it just needs to be right.

That's the more subtle argument for putting AI inside the systems that already hold that context, rather than next to them. We've written about what that actually looks like inside a CRM, but the short version is: the software that wins is the one closest to the data. The type of model seems less important when it comes to real use cases, like CRM.

More AI tools created a new problem: too many places to look

There's a less-discussed side effect of generative AI going mainstream. Every team now has three or four separate AI tools open at once – one for writing, one for meetings, one bolted onto whatever app they were already using – and switching between them has its own hidden cost.

The research on this is pretty stark: constant tool-switching costs the global economy hundreds of billions in lost productivity a year, and it takes over 20 minutes to properly refocus after a single interruption. Adding another AI tool to fix a productivity problem can, ironically, create a bigger one.

The next differentiator is trust

Model capability is converging. What isn't converging as fast is governance – the parts of AI adoption that decide whether a business can trust what the AI is doing with its data.

Gartner's own analysis of the agentic AI market makes this point directly: adoption is accelerating faster than the governance, security, and trust mechanisms needed to run it safely. That gap is especially visible for European businesses, where data residency and compliance aren't optional extras; they're the first question a buyer asks a vendor.

Bottom line: the AI conversation is quietly shifting from "what can it do" to "what is it allowed to do, and who's accountable for it." Data quality is a big part of that answer, as an AI system is only as trustworthy as the records it's reading from.

Where this goes next

Generative AI stopped being the story. What businesses do with the context they already have is the story now, and that's a much more useful question to spend time on than which model is smartest this quarter.

Back to articles