Artificial Intelligence (AI) has led to a tectonic shift in changing the business game across industries and technologies. But like every new technology, opportunities abound as much for improvements as for notorious activities. With its far-reaching effects knocking at our doors, we need to understand where AI is heading, and how we can use this to fuel our growth.
We cover all this in a tête-à-tête with Thierry Caminel, CTO for AI at Eviden. With his extensive expertise in machine learning, generative AI, and knowledge graphs, Thierry has led the AI Expert Community at Atos and overseen projects for clients across various sectors and regions.
We spoke with him to explore the future of AI: Where are we today? What are the key trends emerging on the horizon? And how can businesses prepare for the changes ahead?
Here’s what he had to say.
1) Earlier this year, you highlighted the rapid rise of intelligent agents and published a white paper on their practical implementation. Where do we stand today?
Right now, the entire IT ecosystem is evolving toward autonomous agent technologies.
Large language model (LLM) providers are enhancing their models with functionalities that make it easier to integrate agents: better support for JSON, improved decomposition of problems into sub-problems, and more robust function-calling capabilities. Some are even introducing dedicated accelerators or agents, like OpenAI’s Swarm or Microsoft’s AutoGen.
At the same time, solution vendors are developing agents that enable natural language interaction with their systems, sometimes replacing or complementing RPA or BPM tools. There’s a surge in startups and open-source communities focused on low-code or no-code frameworks designed for these agents.
As expected, agents’ abilities to interact with user interfaces and workstations are also progressing. Some agents now require only a screenshot to understand and autonomously execute user assistance tasks. We’re already seeing this with the latest version of Anthropic’s Claude model, which incorporates such capabilities natively. This evolution is coupled with the emergence of smaller yet powerful language models that can run on smartphones or desktops, many of which are now equipped with GPUs.
Today, models with 8 billion or even 3 billion parameters rival the performance of 150-billion-parameter models from just 18 months ago! This development will enable agents to operate everywhere, communicating and collaborating seamlessly to automate a wide range of human tasks. While still in its early stages, this field is advancing rapidly. We’re witnessing a veritable Cambrian explosion of solutions in this space.
2) What developments do you foresee in the next two to three years? Should we expect new disruptive shifts?
One clear trend is the increasing integration of reasoning capabilities in agents, i.e. the ability to break tasks down into subtasks and use the right tools for each, much like a human would. This will allow AI to handle far more complex requests, such as those requiring strategic analysis or processing vast amounts of documents.
There’s also a fascinating convergence between different types of AI. While deep learning and LLMs excel at pattern recognition, they struggle with true problem-solving. However, combining deep learning with symbolic AI techniques, such as knowledge graphs, can create powerful hybrid solutions. Fields like healthcare and finance, which already have detailed ontologies, could benefit from this quickly. In this vein, DeepMind is making impressive strides. For example, during recent math competitions, they achieved remarkable results by combining LLMs with mathematical proof solvers.
The shift toward multimodality is another strong trend, with AI increasingly capable of processing real-time audio and, soon, video.
Perhaps the most disruptive trend, however, lies in robotics. The hybridization of different AI types will soon enable voice-controlled operation of various robotic tools — from robotic arms to humanoid robots, which could become available for a few thousand dollars. Companies like Tesla, as well as lesser-known players like 1X, Agility Robotics, and Figure, are making rapid advances. Even Hugging Face is now developing open-source frameworks for robotics.
3) How should businesses prepare for these changes? What strategic priorities would you recommend?
Paradoxically, the top priority should be to raise awareness and provide widespread training for employees — including management — on these rapid technological advances and their implications for their roles.
For businesses, having a forward-looking perspective is crucial to anticipating the impacts on solutions, processes, and business models. Being proactive is far better than reacting defensively after the fact.
In our high-intensity tech era, technological foresight needs to return to the heart of strategic planning. Major changes are on the horizon. The productivity gains offered by tools like ChatGPT, Copilot, and Gemini are, in my view, just a ripple compared to the tidal waves of innovation to come.
The impact will go far beyond a simple 30% productivity boost for white-collar workers. We’re already seeing this in software development. For instance, Google reports that 25% of its code is now AI-generated — code that itself supports creating more AI solutions. We’re looking at a feedback loop where AI generates AI to build even more AI!
This kind of recursive acceleration could drive explosive growth. And that’s just one example. The potential applications in chemistry, biology, engineering, and other fields are massive.
We are still at the very beginning of a new phase of disruption, with potentially profound societal consequences. It’s urgent to start understanding and preparing for them.
