The cheap or free use of AI tools is over.

"For the last three years, the most expensive experiment in technology history has run on someone else's money," Fusion5 chief executive Kristy Brown said.

A white paper released by technology services provider Fusion5 sets out the case for a managed approach to the running of AI agents in order to avoid runaway costs, while maintaining compliance with emerging regulatory requirements as well as institutional risks to businesses.

"Who's paying for AI now? Increasingly, it's you," Brown said, adding that many AI agents were adopting a pay-by-token system rather than an all-you-can-eat user licence, or a hybrid scheme which combined a bit of both.

These include Anthropic's Claude, Google's Gemini, Microsoft's CoPilot and many other widely used coding and customer service agents.

"Prior to this phase of our AI revolution, initially it was just about trying to get people to trust and to use AI," Brown said.

She said the challenge now was to use AI more efficiently, with smart processes that improved search accuracy and resulted in fast results.

"Usage-based pricing means every prompt, every reasoning step and every autonomous agent now carries a real, metered cost - and someone must pay it," she said.

"Autonomous agents now commit spend, reconcile transactions and make decisions that bind the organisation. They're also expensive in a way traditional software never was."

She said AI agents reason in loops and can consume an order of magnitude beyond a simple query.

"An agent stuck in a reasoning loop… can generate substantial costs before anyone notices."

She said AI agents needed monitoring and controls or risked landing businesses with a large bill and a nasty governance surprise.

"The businesses that build the discipline to operate AI - not just deploy it - will be the ones still standing when the music, and the free money, stops."

She said the key to improved performance was to collect key business processes in an "ontology", which was industry jargon for a rich, structured description of the underlying business model, including relationships to such things as a company description, the names and titles of senior executives, financial results, and small details like the font and colour of the business brand.

"It's the stuff that sits in your very deep experts' heads.

"And if you take the time to get those things right, it's a fast track way for your AI workloads to be able to get to the end outcome without a lot of these reasoning steps that it has to do if you haven't told it explicitly where to reference that information."

Brown said companies with incomplete data or digitised institutional knowledge were paying to ask AI agents to look for information that wasn't available, beyond the knowledge held by experienced staff.

"Without a shared meaning layer, businesses pay twice: once for the extra compute, and again for the mistake."

She said the transformation was already underway.

Global giants Tesla, Uber, Meta and Amazon were already implementing employee restrictions on AI Agent usage as businesses shift to the traditional model of measuring costs against productivity gains.

"The four biggest US tech firms are on track to spend close to US$700 billion on AI infrastructure this year alone. OpenAI reportedly spends around US$60 billion a year on compute against roughly US$13 billion of revenue. That gap has been a gift to the rest of us," Brown said.

"That gift was being withdrawn."

Microsoft was leading the pack in driving revenue gains through licensing of its AI Agent Copilot, which was an integral piece of its suite of Office tools, according to its latest fourth quarter earnings report.

"This year, Azure revenue surpassed US$100 billion for the first time, and Microsoft 365 Copilot reached over 30 million paid seats, reflecting the confidence customers are placing in us to power their AI transformation," Microsoft chief executive Satya Nadella said.

The case for an agent operations centre (AOC)

Brown said there was a case for a dedicated function that watched every agent in real time, tracked cost per agent with thresholds and alerts, monitored each agent's guard rails, and escalated to a human the moment an agent stepped out of line.

"This is exactly why Fusion5 is investing in building an Agent Operations Centre," Brown said, adding the service would be officially launched in the fourth quarter of the year, with 1400 existing customers in the New Zealand and Australian region.

She said an AOC was aimed at helping small to mid-sized businesses, who don't have the in-house expertise to manage themselves.

"For a business without Silicon Valley's balance sheet, that math matters even more."

She said the proliferation of AI assistants or agents had introduced a class of operational challenges that conventional IT management was never designed to address.

"Business functions that rely on autonomous agents for critical workflows cannot simply revert to manual processes when agents fail or are unavailable," the white paper says.

"The more deeply agents are integrated into value-creating processes, the higher the stakes of operational failure."