Agentic AI is emerging as a critical capability for modern enterprises. Gartner predicts that by 2029, AI will autonomously resolve 80% of common customer services issues without human intervention, leading to a 30% reduction in operational costs.[1]
For enterprises, this is not about replacing people, but upgrading how work gets done, while keeping humans in the loop for oversight and accountability. At the same time, they are enhancing operational resilience, scaling expertise and unlocking new business opportunities and ways of running core functions.
We are on the cusp of the age of intelligence
This promise has attracted a considerable amount of investment – over $400 billion in AI infrastructure in 2025 alone. At $12 billion, the returns have so far been small, and Usman Javaid, Chief Product and Marketing Officer at Orange Business, suggests that we might be in a bubble that could burst. This takes place at Orange OpenTech 2025.
“But what is important is what comes out of that bubble. What we saw with dotcom was the emergence of social media, digital commerce and so on. I think there will be a similar outcome. It is going to be the age of intelligence,” argues Javaid. This is causing a lot of anxiety amongst enterprises, who don’t know which AI train to get on. “What is really important is that they need to make the right choices because they need to get started,” he says.
Regarding the AI bubble, a significant shift is underway, according to Steve Jarrett, Chief Data and AI Officer at Orange. Instead of using enormous amounts of computing power to train models and very little for inference, models are now allocating more compute resources to reasoning through the steps needed to accomplish tasks. “This will ensure that if the bubble bursts, there will still be an enormous amount of value there,” he says.
Learning from the cloud hype cycle
The cloud hype taught enterprises that transformative technologies falter when expectations outpace readiness, governance and real-world business value.
Javaid said that AI can learn many lessons from this, as we often overestimate in both the short and long term when significant technology shifts occur. “When cloud adoption was happening, everyone thought everything was going to be in the cloud. Today, roughly 40% of workloads are not on cloud”, he notes. Not every application is suitable for the cloud; the same applies to Agentic AI. “You don’t need to solve every problem with agents,” he maintains.
To avoid this, Jarrett says the way Orange thinks about agents is “they are best at creating a first draft”, with humans in the loop reviewing and providing feedback. He warns, however, that to make this work at scale, strong tooling is paramount for monitoring, managing, and iterating on the agents across different AI models.
Karine Palacios, Chief Product Officer of Vocalcom, which provides cloud-based omnichannel contact centre software, adds that the value of humans within the use of agentic AI must not be forgotten. “We have been asking AI to collaborate with humans for a while. While agentic AI may be able to deal with more complex processes and conversations. It is still humans that are the heroes of your brand and loyalty,” she says.
“You need to make sure that your brand has the empathy to offer what the tools cannot,” adds Javaid. “You need to assist those humans to be able to do their job better,” citing the fact that agentic tools can reduce the time humans take to find information by 30%, enabling them to spend more time with the customer, creating a better experience.
Making AI more sustainable
Agent-based AI is rapidly gaining traction, with large language models increasingly being trained to perform agentic tasks. Significant innovation is occurring in mid-sized models (7-20 billion parameters), which offers a strong balance between capabilities and efficiencies.
Many of these models are open weight, where the trained parameters are publicly released. This enables sovereign AI solutions that do not depend on public cloud. Orange, for example, has helped a French bank build coding models in the sovereign cloud for agentic AI that would previously have needed to be hosted outside Europe, which would have been impossible due to security reasons.
This open ecosystem is accelerating innovation, explains Jarrett, which demonstrates that organizations do not need to use the largest, most resource-intensive models for every use case. “It enables us to be responsible, choosing the right tool for the right problem – being economically and ecologically responsible,” Jarrett says.
Making a difference to real-world operations
Despite the massive investments in AI, real-world outcomes are the true measure of success. Javaid cites a recent MIT study[2] that found 95% of enterprises have failed to see a measurable return on their investment in AI. He believes that enterprises have seen productivity gains but not the return on investment. “These are the two different expectations,” he says.
Javaid says that Orange has seen use cases that range from job-line cases to industry-specific ones, such as drug discovery and predictive maintenance. He notes that he has identified three main areas driving scale and productivity gains: employee productivity, customer experience and support, and software development, where enterprises have seen up to a 10 times increase in efficiency.
Javaid believes the biggest challenge to the adoption of agentic AI is a trust deficit. “As you see these technologies getting into the hands of bad actors, the trust deficit is further created,” he says. “This is a big responsibility we need to get right”.
Reaching that Henry Ford moment
The real value of AI lies in co-intelligence, where AI works with humans, which underpins the move Orange is making to full agentic AI, explains Jarrett. This approach not only boosts productivity but also helps to close the digital divide for people worldwide. “Being able to solve problems in their daily lives that they would struggle with otherwise,” he says.
For agentic AI to have its disruptive “Henry Ford” moment, however, Javaid said we must rethink systems from the ground up rather than bolting business processes onto AI.
“AI is going to absorb a lot of mediocrity and magnify the excellence – but we still need to make sure that humans hold the reins,” he concludes. “Progress means nothing if it leaves humans behind”.
[1] Press Release: Gartner 2025 Agentic AI is set to transform the customer service landscape https://www.gartner.com/en/newsroom/press-releases/2025-03-05-gartner-predicts-agentic-ai-will-autonomously-resolve-80-percent-of-common-customer-service-issues-without-human-intervention-by-20290
[2] https://fortune.com/2025/08/18/mit-report-95-percent-generative-ai-pilots-at-companies-failing-cfo/
