Beyond the headlines: what AI changes about how IT is run

Most AI coverage is about capability. Less of it is about what changes in the ordinary work of running a technology function. These five are the ones that show up in budgets, contracts and risk registers.

1. Part of the licence bill moved from seats to usage

For twenty years software cost was a seat count multiplied by a price. AI features are increasingly billed by consumption, which behaves differently: it rises with adoption, it can be started without procurement, and nobody is watching it in the first month.

The consequence is structural. Budgets built for fixed costs need a forecast, a ceiling and an alert for anything billed by use. This is ordinary cost management applied to a new line — it just has to happen before the invoice.

2. The threshold for building software dropped

Work that once justified a multi-quarter project now lands far faster, and ideas that could never carry a business case now can. That moves the build-versus-buy line, and it means decisions made a few years ago on the old economics deserve revisiting — particularly where a product was chosen because building was assumed to be out of reach.

It does not mean build more. It means the comparison changed.

3. Your content estate became a security control

An assistant answers from what the person asking is already permitted to see. Nothing new is exposed, but a great deal becomes findable in plain language that previously required knowing a document existed.

That makes permissions, duplication and retention part of the security posture rather than a housekeeping matter. In most organizations this is the largest piece of preparatory work and it cannot be compressed by adding people.

4. Procurement has new questions

Every vendor now has AI features. The questions that matter are consistent: what does it do with your data, where is it processed, is it used for training, can the feature be disabled, and what happens to the pricing when usage grows.

These belong in the standard assessment rather than in a special AI review, because by now there is no such thing as a product without them.

5. Policy became necessary earlier than most planned for

Staff have access to capable tools through the software you already buy, and through their own devices regardless. The decision about what may be put into which tool has effectively been made by default in most organizations.

What is needed is short: what may go in, what may not, who is accountable when an answer is wrong, and how something new gets approved quickly. A page or two, in use, beats a framework in draft.

What has not changed

The things that decide whether any of this works are the same as they were: data people trust, processes that are understood, a clear view of what is being asked for, and someone accountable for the outcome. AI raises the value of all four. It does not substitute for any of them.