The term “agentic AI” is rapidly gaining traction, capturing the attention of both tech enthusiasts and skeptics alike, all of whom are pondering its cost-benefit balance amidst significant global investment. While some question the return on investment from artificial intelligence, data suggests that agentic AI could be a promising solution enabling considerable efficiency gains across various sectors.
Diving deeper into the landscape, last year’s statistics revealed that less than 1% of enterprise software applications incorporated any form of agentic AI. However, a notable shift is on the horizon, with tech consultancy Gartner forecasting that by 2028, this number will rise to 33%. This expansion is projected to allow for 15% of routine work decisions to be autonomously made by AI systems, indicating a transformative trend in workplace dynamics.
Unlike earlier iterations of AI, which primarily focused on generating text or assisting users through queries, agentic AI offers a new framework in which the technology can autonomously perform complex tasks. In a keynote at Turing Fest 2025, Jonathan Field, a solution architect at OpenAI, shed light on the company’s latest Agent releases, showcasing their potential to revolutionise business practices.
Field commenced his address with a striking remark: “You can’t go on LinkedIn without seeing somebody mention [agentic AI]. The reason everybody is talking about it is because there’s a huge shift going on right now.” He elaborated on how the evolution from ChatGPT-style interactions—where users posed questions and received answers—has morphed into a paradigm where AI is tasked to operate independently.
OpenAI’s dedication to investing in advanced agentic AI functionalities is evident in their recent tools. One standout, named Deep Research, demonstrates how AI can undertake time-intensive processes such as market analysis or policy benchmarking. This tool is designed for a “longer horizon,” enabling it to source information, evaluate various documents, and offer structured insights—all without the constant prompting required by traditional chatbots. This level of autonomy is akin to what a junior analyst might accomplish.
Another innovative addition is the Operator, described as a “computer-using agent” that automates web-based tasks by mimicking human interactions. As Field explained, this tool can seamlessly navigate websites, complete forms, and engage with multiple features dynamically. Companies have begun using Operator for tasks like automated usability testing, reducing the burden of manual checks typically required for quality assurance—a perfect illustration of how agentic AI can streamline operations.
Yet, the implementations come with challenges. Questions surrounding data integration and quality, alongside concerns about cybersecurity, loom large, echoing sentiments from various sectors leaning into agentic AI. A broader analysis suggested that while these innovations could lead to substantial productivity boosts and revenue growth across industries such as healthcare, finance, and logistics, there are formidable hurdles to clear. Companies are investing substantial funds into AI agents with expectations of clear returns, although uncertainties in business models persist.
Field’s keynote did not merely serve as a promotional platform for OpenAI; he provided practical guidance for businesses navigating this new terrain. He outlined essential steps businesses should consider before diving headlong into agentic AI projects. First and foremost is scoping, where Field urged companies to consider if there is truly a need for a sophisticated logic system. He advised caution against the common misconception that every requirement warrants a custom-built model, suggesting that simpler, readily available solutions might suffice.
Moreover, Field underscored the importance of starting with single-agent systems rather than jumping into complex multi-agent frameworks that can introduce significant risks. His advocacy for simplicity was clear: “If you have a use case where something works 80% of the way and that’s okay, you should just do that.” This pragmatic approach is echoed throughout the industry as organisations weigh the risks of relying on advanced systems.
Furthermore, Field stressed the significance of thorough pre-launch checks to ensure systems can handle expected operational loads and highlighted the importance of continuous evaluation. He described evaluations as “the single most critical thing” businesses often neglect, essential for ensuring agents perform as intended. “If you skip evals, you’re flying blind,” he warned, underscoring that without constant measurement and oversight, businesses might find themselves enmeshed in unexpected complications.
The narrative surrounding agentic AI is becoming increasingly multifaceted. While some may dismiss it as mere buzzword, the groundwork laid by initiatives from companies like OpenAI indicates that the shift from auxiliary assistants to autonomous operators is underway. As organisations strive to harness the practical benefits of this technology, those committed to measured execution and careful planning will likely be the ones to succeed in realising the potential of agentic AI.
In an environment where capabilities are rapidly evolving and the landscape is rife with opportunity, it is vital for businesses to adopt a thoughtful approach. Starting small and prioritising assessments could pave the way for transformative outcomes, as agentic AI promises to reshape the fabric of operational efficiency in unprecedented ways.
Reference Map:
- Paragraph 1 – [1], [2]
- Paragraph 2 – [1], [3]
- Paragraph 3 – [1], [4]
- Paragraph 4 – [1], [5]
- Paragraph 5 – [1], [6]
- Paragraph 6 – [1], [2]
- Paragraph 7 – [1], [4]
- Paragraph 8 – [1], [3]
- Paragraph 9 – [1], [5]
- Paragraph 10 – [1], [7]
Source: Noah Wire Services
Subscribe to Industry Updates
