Jens Bontinck, until recently an engineer at Belgian AI company ML6, thinks the future of AI does not lie with general models but with AI agents. “Companies that do not keep up with this development risk falling behind the competition.”
In this parting interview with Techzine, held shortly before he moved to a spin-off from ML6 within the same holding company, Bontinck shares his vision of the role AI agents will play as far as he is concerned. He also explains how ML6 is helping companies implement this technology.
“What you see happening today with language models like ChatGPT is that they are getting bigger, better, more powerful, and more capable,” Bontinck explains. The catch is that such models can be pretty costly and don’t necessarily have (or should have) access to proprietary information.
Everyone knows the stories of employees who—sometimes without management’s knowledge—start querying general LLMs about work processes, perhaps even entering sensitive or confidential data in the process. This is a compliance nightmare and not even very effective regarding the intended goal.
This is where specially applied AI agents come in. “These can work effectively by specializing in one domain, using information from the CRM system, for example. In this way, they specifically consult a company’s existing knowledge database, while that info remains secure on its own servers or cloud environment.”
Making large models smaller
This approach allows deploying smaller, specialized agents instead of huge language models. “The known large models remain very useful of course, if only because they can form the basis for smaller, specific agents. We can downsize and fine-tune those LLMs to create a very specialized agent,” Bontinck said.
Ideally, it won’t just be one agent. Bontinck believes in the power of multiple specialized agents working together on one project: “A hundred agents are going to perform better than a hundred generalists.”
He also emphasizes the technical advantages of this approach: “Such a large language model doesn’t work on an iPhone or on a device with a small memory. For our work, we regularly program an assembly line in a factory. Such a device often only has one, two, or at most four gigabytes of RAM. We shouldn’t even think of running a local ChatGPT model there.” A specialized agent will be much more helpful in such cases.
Going all-in on AI agents
ML6 has a lot of expertise in AI and machine learning applications. The company had already been working with it before the current transformer technology (the ‘T’ in ChatGPT) accelerated artificial intelligence, but it has since put plenty of effort into the latter development. It helps organizations develop and implement AI solutions and has built a client base in healthcare, retail, manufacturing and financial services, among others.
As a vanguard in the AI revolution, the company sees great promise in applying agentic AI, or AI that can act and make decisions partly autonomously. The idea is that such agents don’t just respond to predetermined instructions or set patterns but can set their own goals within certain parameters, initiate actions, and possibly adjust them based on feedback or changing circumstances. They can all have separate roles that make agents manageable and affordable. So, individual agents don’t necessarily have to be all-knowing and all-powerful.
Tip: Your Next Colleague Is an AI Agent; 2025 Will Be Year of Agentic AI
ML6 is working on several projects where AI agents play an important role. In customer support, for example, agents can help route questions and answer them based on an internal knowledge base. Bontinck gives a concrete example: “An agent can make sure that a question goes to the right service. It is a simple example, perhaps, but a lot of time goes into this for people. Who is the question for? Is it for finance? Legal? Should it go to the complaints department? Or rather to customer success?”
Service-as-a-software
“A next step would be not only to forward those questions to the right department but also to have a knowledge base from each department that allows the agents to answer customer questions right away. In that way, AI agents are the hands and feet we give to LLMs.”
According to Bontinck, this leads to a new definition of SaaS: “In the future, SaaS will no longer mean software-as-a-service, but service-as-a-software. Powered by AI agents that can offer a full suite of those services.”
There are plenty of other use cases besides customer service. ML6 is also working on a project called Decipher, where they are deploying AI in biology. “We’re trying to manipulate the DNA of a bacterium so that it can generate molecules to counteract certain disease states,” Bontinck says. “At the same time, we want that living bacterium not to die, and to stay in the lab environment.” This project, in collaboration with the University of Ghent and others, shows how broad the applications of AI agents can be.
The company’s customized AI solutions range from automated decision systems to predictive analytics. ML6 (the name is a play on Britain’s foreign security service MI6) also works with deep learning algorithms, natural language processing (NLP), and computer vision.
There is enough business to go around, you might say. Yet the company believes it is important not to wait for clietsn to knock on the door with an AI question. Bontinck: “We are proactively identifying use cases in different sectors. We try to bring our own story to industry and clients using our experience and insights.” Potential customers often don’t know what they need, and ML6 is happy to explain that to them.
Educating clients
Recently, ML6 has set its sights on areas such as corporate mergers and acquisitions, an industry in itself. Bontinck says there is also lots of potential AI business in product innovation, procurement processes and ESG reporting. “We try to really actively question targeted clients in customer surveys and also provide a bit of guidance,” Bontinck said.
He compares this approach to Henry Ford, who once supposedly said, “If I had asked people what they needed, they would have said a faster horse.” Bontinck: “No one knew what a car was yet. It’s similar to our services, as potential clients often don’t yet know what they need.”
Bontinck emphasizes the importance of this proactive approach: “Of course, it’s a way for us to gain clients, but we also see it as our duty to educate, to show what can be done. If we can improve the legal business or the procurement sector in the process, I am proud that we can. It’s a bonus for us to add many new use cases to our portfolio that way.” Bontinck refers again to the biology project. “When we can say to have participated in a project that makes it possible to prevent diseases, that’s exactly the difference we want to make.”
Ignoring AI is not really an option
According to Bontinck, implementing AI is a no-brainer. So the question is not if, but how. “It’s not really a choice anymore whether you are going to do AI or not,” he states. “The saying goes either you compete with AI or against AI. That’s the choice businesses need to make. If you are an absolute market leader and completely dominant in your field, you might have the luxury of ignoring AI. But for most companies, that’s not an option.”
“Businesses might think AI is too expensive so that they won’t invest in it. Well, everyone around them is making that investment. So, the competition will boost their sales and, on a national level, even add to the GDP using AI. Eventually, their whole industry will tank, and the economy might go into a recession. I’m exaggerating a bit, but we must be attentive to the investment versus the return. Yes, AI costs money. But it will pay off in the end. If only because it lets you keep up.”
‘AI will become part of business process’
Bontinck sees a future where AI has become an integral part of business processes: “That won’t take another five years. I believe very strongly that we, as humans, are programmed to do better every time. So, every company will ask itself: how can we start using AI, especially agents? From an innovation perspective, or out of pure necessity in terms of competition or cost.” Companies that do not keep up with this development, he says, run the risk of being left behind.
He expects that in the coming years many companies that are not yet on board with AI will make that switch anyway under market pressure. “They think: Hey, our competitors are doing something with AI. We need to spring into action.” So what does that mean, other than of course calling ML6? Mainly that companies need to get their data right, because without good data, there is no good AI.
So Bontinck expects a lot of investment in data management to support the use cases discussed. Not so much to keep feeding LLMs mountains of general quantitative data (‘That work is done at some point’), but for quietly unlocking ever more focused, high-quality data for the agents who can work with it.
Giving most challenging tasks to humans
Although AI is going to take over many tasks, ML6 wants to be careful not to “automate processes to death. “In a vast majority of jobs, AI is going to give augmentation—not automation, but augmentation, or doing your job even better tomorrow thanks to AI.
“You can’t just trust AI with everything, despite the advances happening almost weekly. AI models hallucinate; I don’t expect that to change anytime soon. What happens when a language model like ChatGPT talks nonsense? What should we do then? That’s where people come in.”
Bontinck sees the partial outsourcing of tasks to AI as an opportunity: “It could just be a chance to give the most difficult and challenging cases to human beings and automate the boring, repetitive tasks.” He acknowledges this is a personal issue: “One could say, yes, AI threatens my job. One could also see it as an opportunity to learn something new.” He does acknowledge that a small number of jobs will disappear because of AI. “The people that this affects need be offered retraining.”
Red lines
ML6 says it understands such concerns and takes the ethical aspects of AI seriously. Bontinck: “We at ML6 have defined several red lines. Take defense applications for example, causing real harm to real people, as a rule we are not going to do that. Beyond that, we want to be transparent at all times and ensure it’s all secure and compliant .” The company has a rigorous process for evaluating projects: “We have qualification meetings to discuss all our deals. And that qualification meeting always includes our head of security, as well as our ethical lead.”
It can lead to specific clients being rejected. “We don’t always have that luxury, and it’s a gray area. Usually, a solution is not just good or bad. But we try to look at this carefully on a case-by-case basis,” said the engineer.
Getting hands and feet dirty
ML6 sees itself guiding in the ongoing AI revolution. “It really is a revolution, one we want to steer, preferably a positive direction,” according to Bontinck. The company deliberately focuses on sectors where they can have a positive impact, such as biology and energy.
With offices in Ghent, the Netherlands, and Berlin, ML6 strives to stay close to the customer. “What we really want to avoid is staying walled off into our ivory tower and explain AI from above. Because it can have such a big impact on companies, we really want to be on the ground, getting hands and feet dirty. Those of the LLMs, but also our own.”
We’ve published a lot about AI agents recently. Read more here:
- Salesforce raises the AI bar with Agentforce; are AI agents replacing employees?
- ServiceNow takes self-service experience to next level with AI agents
- Google launches AI agents in customer engagement suite and Gemini comes to Workspace
- HubSpot deploys specialized AI agents on platform
- SAP copilot Joule can now handle complex tasks