Gartner declares ‘agentic AI’ the next step function

The ‘internet of agents’

Brethenoux’s seven trends pointed to the move away from monolithic models towards complex, multi-faceted systems.

These include composite and neurosymbolic AI that combines neural networks with symbolic logic to improve automated reasoning, transparency and reliability though integrating models that degrade at different rates remains difficult.

There’s also world model AI that focuses on building systems with a holistic understanding of physical reality, expanding beyond domain-specific language models to reduce hallucinations, as well as first principles AI, which combines machine learning with scientific laws so that “even if something has not happened before, the equation can catch it”.

Perhaps the most disruptive trend is what Gartner calls adaptive collective AI, or the internet of agents. This involves distributing decision-making across multiple systems to solve complex problems. Brethenoux cited a Gartner client currently using a swarm of drones to inspect wind turbines in the North Sea. The drones independently photograph cracks and colour degradation, and on returning to base, they collectively decide if there are issues and autonomously generate a report.

Despite these advancements, Brethenoux believes there is still a role for humans. Quoting child psychologist Jean Piaget, who said, “Intelligence is what you use when you don’t know what to do”, he pointed out that generative AI merely replicates things that have been done before, rather than developing new ideas.

The metadata explosion

According to Mark Beyer, Gartner research vice-president and distinguished analyst, investment in AI-ready data is now the top priority for data management executives.

However, preparing this data creates another challenge: “Every time you reuse data, 100 new metadata points are created about that data,” Beyer explained. “If you access that data 100 or 1,000 times more frequently, you’ll be overwhelmed with metadata in a matter of weeks.”

Because humans cannot keep pace with the exponential increase in metadata, Beyer argued that agentic AI is inevitable in the data ecosystem. AI agents, from simple task-based entities to complex multi-agent swarms, will soon handle connectivity, orchestration and data governance.

These autonomous processes will fundamentally change data engineering. For instance, an AI agent could notice that data quality consistently degrades on Tuesdays and automatically appends a warning note to reports generated on that day.

Beyer’s final takeaway to the Sydney audience was that organisations are no longer building data pipelines for analytics or AI. “Rather, we’re now building agents that recognise how the data is used, how often it is used, which part of the organisation uses it and, most importantly, whether it leads to the desired outcome,” he said.

Applying manufacturing principles to FinOps

As agentic systems scale, so do the costs. Adam Ronthal, Gartner research vice-president, suggested that data management needs to adopt the ruthless efficiency of Henry Ford’s automobile production line, where every step has a measurable cost and output.

“I’ve been thinking quite a lot in the last year about what the impact of AI and agentic capabilities is on cost optimisation,” Ronthal said.

While calculating the exact cost of a specific SQL query or report is a “largely solved problem space” through modern FinOps practices, accurately measuring the value of that data remains a major challenge. To solve this algorithmically, Ronthal urged organisations to look at empirical evidence.

For example, if a set of dashboards is used every day by a lot of people, they are clearly valuable. “Frequency of access is a strongly correlated proxy for value,” he said.

By contrast, if a good portion of users are downloading significant amounts of data from a data warehouse into their own analysis environment, they are not getting the value they want.

“We can make the argument that value is derived from a couple of core components: frequency of access [the times when something last ran], and how important it is,” Ronthal said.

By using metadata to understand consumption patterns, organisations can rank their workloads. Once workloads are empirically ranked by value, the enterprise can deploy an agentic optimisation framework. If a critical report is needed by 9am every day, an AI could deploy agents to select the most cost-effective cloud compute instances, structure the data and guarantee the service level agreement without human intervention.

To prepare for this automated future, Ronthal called for IT leaders to start investing in metadata – especially from log files – to truly understand AI consumption patterns. “Over the next 12 months, start to pilot the concepts of empirically derived value and ranking by value. If the number of arguments over relative value decreases, you know you are making progress.”