Here is a slightly unsettling thought to start your week. While you were reading the headlines about AI, AI was reading your competitor’s customer data — and quietly running parts of their business.
This is not a prediction. On 26 June, the news made it official: agentic AI is no longer a slide in a strategy deck. It is being switched on right now, inside real companies, handling advertising, customer service, content and media buying. The same week, OpenAI opened a preview of its new GPT-5.6 models. The tools are here, they can take actions on their own, and they grow more capable by the month.
So that is the scary half. Now the comforting half — and it is bigger than you think.
The unsettling bit
An “agent” is AI that does not just answer questions. You give it a goal, and it plans the steps, uses your tools, and gets the job done. Pointed at the right work, that is genuinely powerful. The worry is simple: if it is acting on your behalf, it is acting on your data — and most companies’ data is a mess.
A global survey of marketing leaders, published the same day, found that only one in ten rate their organisation’s technology as truly ready. Ten percent. Everyone else is bolting fast, autonomous AI onto shaky foundations. That is the real risk — not killer robots, but confident software making the wrong call, quickly, at scale.
The comforting bit
You do not need to win an AI arms race. You do not need the newest model the day it launches. What you need is far more boring, and far more in your control: data that an agent can safely act on.
That means one clear view of each customer instead of five scattered records. Clear rules about what an agent may do alone and what needs a human to approve. Data named and organised well enough for a machine to follow. And consent you actually trust. None of it is glamorous. All of it is achievable.
Here is the quietly reassuring truth: the companies that win with agentic AI are not the ones with the flashiest tool. They are the ones whose data was ready when the tool arrived. That is a race you can win at your own pace, starting today.
What to actually do this week
Do not rush to deploy an agent. Do the opposite. Pick one small, low-risk task you would happily hand to a careful junior. Look honestly at the data behind it. If you would not trust a new starter to act on that data without checking, an agent should not either. Fix that one patch of data first.
Agentic AI being switched on everywhere sounds alarming. But it just turns up the volume on something we have always believed: get your data right, and the scary new technology becomes a quiet advantage. Get it wrong, and the speed simply makes the mistakes louder.