Copyright Fast Company
Large-scale, general-knowledge models might be the current face of AI, but it’s the smaller, more targeted systems that are quietly driving the AI revolution. Enterprise efforts to incorporate AI into workflows face challenges in rising costs, security concerns, and the potential for hallucinations. In the face of these concerns, the type of model deployed makes a difference. Scaled-down, more customized solutions are increasingly giving organizations the security, efficiency, and accuracy they need to move ahead with confidence. At the 2025 Fast Company Innovation Festival, IBM sponsored a panel discussion about the advantages of smarter, more specialized AI models for business applications. The conversation among three experts with diverse industry experience highlighted the ways a more modular approach to AI systems can improve security, produce better-tailored solutions, and meaningfully improve the lives of the people using these tools. Here are three key takeaways. (Some quotes have been edited for length and clarity; scroll to the bottom to watch the entire panel discussion.) 1. A modular approach creates more efficient tools. AI models may be fun to play with, but businesses need them to be reliable and accurate. A general-knowledge AI engine has to know something about basic physics if it’s going to help a high-school student understand a homework assignment. For an HR chatbot, on the other hand, that information is overkill—and it leads to inefficiencies. A popular misconception about AI imagines a single sprawling model that does everything. But today’s AI is more like software, explained David Cox, VP of AI Models at IBM Research and IBM Director of the MIT-IBM Watson AI Lab: “a lot of different parts put together in an intelligent way.” This level of optimization is great for businesses, Cox said, offering “an amazing toolbox of different [model] vendors that we can pick and choose from.” 2. Managing risk is essential. AI models can create serious risks for businesses. The more guardrails organizations can erect to protect against these risks, the more widely they can deploy AI tools. Safety measures include protecting the system itself from hackers, keeping data secure, and preventing models from producing output that could be wrong, costly, or damaging to a company’s reputation. The use of open-source tools is helping to make AI safer by getting more eyes on potential problems and taking advantage of a more diverse set of experiences to inform best practices. “If different kinds of people are building the technology, then the technology will not have as many of those blind spots that we sometimes see,” said Cox. Eliminating the blind spots is important, as AI tools ultimately serve a purpose that often has to be easily interpreted by a non-technical user. “If it’s not usable, no one cares if it’s great or not,” said Ben Colman, CEO and cofounder of Reality Defender, a cybersecurity company that helps organizations defend against deepfakes. “Our biggest challenge is boiling down complex output into a ‘yes’ or a ‘no’—a green check mark or a big red X.” advertisement 3. AI tools should be designed with people in mind. Having a modular set of tools available effectively democratizes the development process for AI solutions. That’s a good thing for their overall effectiveness, and it increases the chances of buy-in from those who will actually use them to improve their productivity. It also addresses a common fear among employees that AI tools will take their jobs. “We want to teach people that their jobs can be easier,” said Chelsea Kaden, chief people officer at eyewear retailer Warby Parker. “We’re really focused on getting our corporate teams to leverage tech like AI to make the client-facing teams’ jobs more frictionless—when you’re in front of an angry customer, you want to answer their question right away.” Getting buy-in leads to success, which leads to enthusiasm. That starts a virtuous cycle in which the people who will eventually use these tools have a role in the innovation cycle, helping to identify additional situations where AI could make their jobs easier. “The more team members can do with the same tools they use every day, with the added benefit of AI, just makes them that much stronger,” Colman said.
