AI in Manufacturing: What's Now—and What's Next?

The rise of artificial intelligence signals a revolution in technology. But it's actually even bigger than that. In many ways, it marks the beginning of a transition to an entirely new economy—one in which highly complex, customizable products will be manufactured on-demand using intelligent manufacturing systems.

The truth is, AI is still something of a moving target, and it can be difficult for business leaders to separate hype from reality. But in manufacturing, the timely adoption of AI workflows could spell the difference between failure and success, making it critical for leaders to stay on top of emerging trends and understand exactly what is possible today.

Engineers often come to my MIT Professional Education course on computational design and manufacturing with only a hazy idea of current AI capabilities. But after a mix of labs and lectures, they leave with a clear understanding of how AI tools are already having an impact on manufacturing, where things are headed in the coming years and what they can expect to see over the next decade as this manufacturing transition plays out.

Today: A Useful Starting Point

Already, technology is enabling design automation, with generative AI platforms generating CAD code from simple text prompts that outline basic design parameters. For example, you can ask a model to design a chair, and you'll get working code for a 3D model within seconds.

The results still aren't perfect. You'll likely need to adjust the alignment of certain components or make modifications to some aesthetic elements. But this process gives engineers a useful starting point that they can build upon through additional prompts or manual interventions.

Neural surrogates—machine learning models trained to replicate physics-based simulations—are now replacing weeklong finite element simulations with predictions delivered in under a second – achieving 10-17x speedup while maintaining over 95% accuracy, according to recent aerospace and nuclear applications. We're beginning to see more companies build out custom versions of AI tools that are trained on their own data, which is leading to even more accurate and useful designs.

Just as importantly, AI is taking over rote, manual tasks (such as documentation) that engineers typically hate doing. This eliminates much of the "boring" work of engineering, allowing engineers to focus on solving interesting and challenging problems. Of course, harnessing this power requires navigating significant challenges, from ensuring the security of proprietary design data in AI models to managing the integration complexities of legacy systems. For now, AI design solutions still require significant human oversight – but these tools are significantly accelerating and streamlining the work of those human engineers.

Tomorrow: More Prevalent in the Everyday

Although AI tools have improved rapidly over the past couple of years, many companies have found themselves stuck in the experimentation phase. Almost every organization has toyed with interesting use cases, but not everyone has discovered a practical path forward for bringing these use cases into full production. According to McKinsey's 2024 research, 78% of organizations now use AI in at least one business function, up from 72% earlier in the year, yet most struggle with scaling beyond pilots.

Over the next year or two, we will see AI become much more prevalent in the everyday operations of manufacturers, with the technology powering instant customization of products and automated error detection for new designs. Much of this improvement will be achieved through integration of various AI tools. We're likely to see more specialized platforms pop up for design, simulation and manufacturing—and then, we're likely to see these disparate tools connected within unified systems.

Rather than manually transferring their project data between different software environments, manufacturers will work within AI software suites that bring together multiple workflows. While a handful of organizations are already creating custom models trained on their own data, these will become the norm over the next 18 to 24 months, allowing companies to maintain control over their sensitive intellectual property data as they improve their design and manufacturing processes – addressing one of the primary regulatory compliance concerns around data sovereignty and trade-secret protection in AI-driven manufacturing environments.

Ten Years from Now: Building on Demand

In the future, we may see a scenario where consumers themselves will be able to describe products they need, then watch as AI systems build those products for them on-demand. Without reaching too far into the realm of science fiction, it's easy to imagine someone asking for—and receiving—a robot that will complete their yardwork, built specifically to handle a quarter-acre parcel with three mulched flowerbeds.

It is quite likely that these customized products will have capabilities that extend far beyond what is possible today. Rather than simply mowing and trimming, for example, our hypothetical lawn-care robot might analyze soil conditions, make adjustments based on the weather and optimize cutting patterns for different grass types. This sort of automation is going to come in waves. The automated production of simple consumer products will happen years before Lockheed Martin applies similar processes to the F-35 fighter jet.

As manufacturing technology evolves, we will still need human engineers, but their roles will be very different. Engineers will still maintain responsibility for signing off on final designs, but instead of creating these designs from scratch, they may largely spend their time verifying and approving design work performed by AI. The challenge for manufacturers will be to embrace the innovation offered by AI, while also enforcing the exacting level of precision required of product engineering