This is one outlet's own report from TechRadar — the article as it was filed.
AIPROPX ReportTechRadar · 1h ago
The next AI phase is better agents not bigger models
For organizations looking to gain an edge by adopting AI tools , the decision has been dominated between large language models.
Every new release promises better reasoning, greater accuracy and more capabilities than the last.
But as foundation models become increasingly powerful and widely available, the real differentiator for organizations won't be which model they choose - it will be how effectively they deploy agentic AI to solve real business problems.
At the recent 2026 Gartner Data and Analytics Summit, the message from analysts was clear; standalone AI models are outdated and autonomous, interconnected AI agents are the future. But to get real value from agentic AI, IT leaders should prioritize “high-frequency, low-complexity use cases, apply guardrails and upskill the workforce.”
To this statement, I would add that the next phase of enterprise AI isn't about having access to the biggest model. It's about building better agents.
That means moving beyond seeing AI as simply another chatbot and instead treating it as an operational capability that can augment employees, automate repetitive work and unlock efficiencies throughout the organization.
The biggest AI opportunity isn't conversational
For many people, AI still means asking ChatGPT a question and receiving an answer. That's certainly one application, and conversational interfaces have played an important role in making AI accessible. But that’s only scratching the surface of what's possible.
Some of the most valuable agentic AI deployments are focused on operational work rather than conversation. For example, in higher education, AI agents are helping students navigate support services by answering routine questions about enrolment, campus facilities or administrative processes, freeing staff to deal with more complex issues.
In transport, agents can combine timetable information, onward travel options and customer support into a single interaction, saving users from having to search across multiple services.
Elsewhere, organizations are using AI agents to qualify sales enquiries, summarize complex documentation , and extract information from contracts. Tasks that previously required hours of manual effort.
The common thread isn't simply that these agents can answer questions. It's that they remove administrative burden and enable people to focus on work that requires judgement, expertise and human interaction.
Better agents start with better problems
One of the biggest misconceptions surrounding agentic AI is that organizations should begin by deciding where they want an AI agent.
In reality, for organizations looking to develop their AI offering, the first question shouldn't be, "Where can we deploy AI?" It should be, "What business problem are we trying to solve?"
Starting with technology often leads organizations to build agents that deliver little value because they're automating the wrong process. The organizations seeing the strongest returns are those identifying operational pain points first, in areas where employees spend large amounts of time on repetitive administration, searching for information or manually moving data between systems.
When those challenges are clearly understood, agentic AI becomes an effective operational tool rather than technology looking for a use case.
The model is becoming less important than the implementation
Today's leading AI models are improving at an extraordinary pace. They're becoming more capable, faster and increasingly cost-effective which means that the competitive advantage is shifting
Organizations are unlikely to outperform competitors simply because they've selected one foundation model over another. Instead, success will increasingly depend on everything that surrounds the model.
Data quality is one of the biggest differentiators. An AI agent is only as good as the information it's grounded on. Contradictory documentation, inconsistent policies, or poor knowledge management inevitably reduce the quality of outputs.
Prompt engineering is another often overlooked capability. Designing an agent isn't simply a matter of giving it instructions and expecting consistent results. Well-designed prompts establish clear guardrails, define the agent's responsibilities and ensure it stays focused on its intended purpose.
The same is true of governance. A good agent shouldn't answer every question it's asked. It should understand its role, know when to decline requests that fall outside its remit and operate only with the information and permissions it genuinely needs. Applying the principle of least privilege isn't just good security practice, it also improves efficiency by reducing unnecessary processing and keeps costs under control.
Ultimately, the quality of the implementation matters far more than the size of the underlying model.
Trust is built through iteration, not ambition
Another common mistake is assuming that agentic AI should be deployed through large-scale transformation programs. But, in practice, the opposite is usually true.
The organizations achieving meaningful outcomes tend to start with a tightly defined use case that delivers measurable value. This creates confidence among employees , demonstrates return on investment, and provides an opportunity to refine governance before expanding into more sophisticated workflows.
Agentic AI should be treated as a product that evolves over time rather than a project that's complete on launch day. However, that also changes the way organizations approach testing.
Traditional software is designed to deliver identical outputs every time. AI agents are probabilistic by nature, meaning consistency becomes just as important as functionality. Continuous testing, monitoring, and refinement therefore become integral to successful deployment.
Just as importantly, employees need to be part of the journey. Building trust isn't about unveiling an AI agent once it has been completed. It's about involving t...
AIPROPX is an independent multi-source news index — we track, compare, and connect coverage from across the web into one place you won't find anywhere else.
