The AI stack you didn't mean to build
AI is not arriving as a project. It is arriving as behaviour.
Ten people. Fifteen tools. No-one planned it.
A couple of personal ChatGPT subscriptions. Someone else prefers Claude. An AI note-taker quietly becomes part of meetings. New AI features switch on inside the CRM and docs stack. Somebody starts trialling an open-source model they saw online. No formal roll-out. No clear operating model. Just a growing business slowly realising AI is already threaded through how work gets done.
That is the pattern.
Not deliberate adoption. Not a clean strategy. Not even a conscious decision, most of the time.
It just starts happening.
And by the time leadership notices, the business often already has a mix of personal AI habits, team workflows and system-level automation beginning to form. That is why so many growing businesses feel both behind and slightly out of control at the same time.
The external signal is hard to ignore:
- Microsoft and LinkedIn found that 75% of knowledge workers already use AI at work, and 78% of AI users are bringing their own tools. In smaller businesses, that rises to 80%.
- Netskope found that 98% of organisations are already using software with embedded generative AI features, while 72% of enterprise genAI use still happens through personal accounts.
- Gartner says 69% of organisations either suspect or have evidence that employees are using prohibited public genAI at work.
The detail varies by source. The direction does not.
AI is already inside the building.
The real problem is not adoption
The market has names for bits of this: shadow AI, AI sprawl, agent sprawl.
Useful labels. But they still miss the deeper point.
The real issue is not simply that people are using unapproved tools. It is that work itself is being reshaped before the business has a clear model for where AI belongs.
That distinction matters.
If you think the problem is "which tool should we choose?", you will try to solve it with procurement.
If you think the problem is "people are using AI without permission", you will try to solve it with policy.
But if the real problem is that AI is seeping into the business through multiple paths at once, then the job is different. The job is to notice it, sort it, and shape it.
This is a familiar business pattern, just faster
We have seen this before in other forms.
Shadow IT happened because people needed to move faster than the official systems allowed. BYOD happened because staff brought their own devices to work before policy caught up. SaaS sprawl happened because useful tools were cheap, easy to trial and locally rational, until the whole stack turned messy.
AI is following the same broad pattern. Microsoft explicitly frames this as the hard part of a familiar technology disruption, and Netskope says genAI adoption in the enterprise is following the typical pattern of new cloud services: people use what helps them now, and governance arrives later.
What is different with AI is the speed.
AI is not only something people open in a browser tab. It is also something vendors are quietly embedding inside the software businesses already run. That means leaders can think they have not "adopted AI" while AI is already influencing meetings, writing, research, support, CRM records and internal workflows.
That is why this catches growing businesses off guard.
A better frame: three lanes, three jobs
Most businesses get stuck because they are trying to make one decision about what is actually three different things.
A better mental model is this:
1. Personal productivity
This is AI that makes one person faster.
Drafting. Research. Summaries. Brainstorming. Analysis. Coding. Cleaning up a messy thought before it turns into a useful piece of work.
This is where most adoption starts, because the value is immediate and personal. Someone finds a tool that helps, and they keep using it.
That is not necessarily a problem. In many cases, it is where the real upside begins.
The mistake is trying to over-standardise this layer too early.
Give people room to become better operators. Put sensible guardrails around privacy, client data and review. But do not confuse personal productivity with a whole-of-business platform decision.
2. Team productivity
This is AI that changes how people work together.
Meeting notes. Shared research. Collaborative drafting. AI-assisted hand-offs. Shared prompts. Small internal workflows that reduce friction across a team.
This is where AI stops being just a personal edge and starts changing coordination.
It is also where things become sticky. Once a team relies on AI summaries, AI-generated briefs or shared assistants, the business is no longer dealing with a few individuals experimenting. It is dealing with a new workflow.
This layer needs design, not drift.
Not a huge transformation programme. Just clarity. Where does shared context live? What gets checked by a human? What becomes part of the team's standard way of working?
3. Business automation
This is AI inside the way the business runs.
CRM. Support. Operations. Finance. Compliance. Internal systems. Department-specific tooling. Vendor platforms with embedded AI. Purpose-built automation tied to real process.
This is where standardisation matters most, because this is where AI starts affecting consistency, accountability and repeatability.
This layer is not about giving everyone a shiny new toy. It is about deciding where automation belongs, where human judgement still matters, and which systems are important enough to govern properly.
Different lane. Different job. Different management move.
Where growing businesses get caught
Growing businesses sit in the most awkward spot.
They are no longer small enough to stay completely informal. But they are not large enough to have endless management layers, procurement controls or architecture discipline.
So what happens?
One person solves a problem locally. Another copies the behaviour. A vendor adds AI into an existing platform. A team starts depending on an AI output because it saves time. Then, without anyone really deciding to, a patchwork starts becoming the operating model.
That is why this feels so slippery.
It is not one big decision. It is a hundred small ones.
The opportunity and the trap
The opportunity is obvious: genuine leverage is being created at the edge of the business.
People are saving time. Reducing admin. Producing better first drafts. Moving faster. Exploring ideas sooner. Removing friction from annoying bits of work.
You do not want to kill that.
The trap is invisible accumulation.
What starts as a few useful habits can harden into fragmented workflows, messy data boundaries, inconsistent quality, duplicated tools and quiet dependency on systems nobody has really thought through. Gartner's warning on shadow AI, alongside Netskope's evidence on embedded AI and personal-account usage, points to exactly that risk: organisations are often only noticing the problem after the behaviour is already established.
That is the line growing businesses need to watch.
Not experimentation versus control.
Experimentation versus sprawl.
A simple way to get ahead of it
Here is where to start.
Do not start with "Which AI tool should we buy?"
Start with these four questions:
- What is already in use? Personal tools, team tools, embedded AI features, all of it.
- Which lane does each use belong in? Personal productivity, team productivity, or business automation.
- Where do we want freedom, and where do we need standards? These are not the same thing.
- What absolutely cannot stay ad hoc? Client data, regulated workflows, core systems, anything customer-facing or operationally critical.
That alone will get most growing businesses further than another month of vague AI chatter.
From there, the shape becomes clearer:
- Personal productivity: allow exploration, with guardrails.
- Team productivity: choose where shared workflows live.
- Business automation: standardise deliberately and govern properly.
Or more simply:
Do not standardise curiosity. Standardise operations.
That is the balance.
The real starting point
Most growing businesses do not have an AI strategy problem first.
They have a visibility problem.
They cannot shape what they have not yet named.
So the first job is not picking a winner. It is seeing the stack that is already forming inside the business. Then separating what should remain personal, what should become a team workflow, and what belongs in formal systems.
That is the real starting point.
Because the question is no longer whether your business will adopt AI.
It already has.
The question now is whether you are going to leave that to chance.