If AI was already the buzzword of last year, end-of-year predictions outside the manufacturing world hinted at the next frontier of generative AI: AI agents. Fast forward to 2025, and this prediction is becoming reality — AI agents are on the rise. While several industries have already embraced this disruptor, the legitimate question for the additive manufacturing sector is: how will it adapt to this shift? The article below answers this question with key contributions from software developers Authentise and Synera.
The constant hype around emerging technologies has frustrated more than a few people — especially when those technologies failed to deliver the promised outcomes. I felt that same frustration last year after reading a Forbes article listing AM among the most overhyped innovations. In hindsight, I know the trade press, myself included, has often amplified this hype through bold headlines. Today, AI — and particularly “AI agents” — is experiencing a similar buzz. That’s why I believe we should approach this topic with caution, focusing not on the hype but on what it truly means for AM users.
And this starts with defining what AI agents are in the context of additive manufacturing.
The general understanding of AI agents is that they are intelligent software systems that can autonomously perform tasks and make decisions to achieve specific goals.
In the context of AM, “everyone has a different answer to that. From Synera’s perspective, it’s something driven by language models with workflows created by humans, but that still requires human feedback,” Andrew Sartorelli, Product Management & Partnership Lead at Synera, told 3D ADEPT Media from the outset.
“We’ve been experimenting with this topic for a while, ever since our 3DGPT release. The current conclusion is that AI in general has three use case buckets:
- Design/Sim agents: think distortion prediction, support optimization, etc.
- Process/QC agents: predictive maintenance, build monitoring.
- Workforce augmentation: this is the big one: Technical Data Package – TDP generation (like ThreadsDoc), AR guidance, root cause explainers.
To us, this last one has the most promise and is accelerating fastest because it’s best at leveraging LLMs and is also the area where autonomous/semi-autonomous actions (the definition of an AI agent) are the most likely to be accepted,” Andre Wegner, CEO and founder of Authentise explains.
An analysis from IT company Fujitsu reveals that in the manufacturing sector, AI agents hold the potential to dramatically improve productivity, reduce costs, and alleviate labor shortages. A similar argument was presented during our conversation with Ivan Madera, Founder of Adapativ AM (formerly CEO of Morf3D).
According to Fujitsu, AI agents typically follow a 4-step process: Perception, planning, decision-making, and execution. In the manufacturing in particular, “AI agents go beyond simple automation. They are expected to play an increasingly pivotal role in optimizing workflows and driving competitive advantage.”
At the research level, a few projects can help assess the potential AI agents may have:
Researchers explored the use of Multiple Large Language Models (LLMs) to monitor and control 3D printing. Each LLM acts as its own intelligent agent, to manage different aspects of the 3D printing process.
These would be specialized LLM Agents. Each agent is focused on a specific task, like detecting print defects, optimizing settings, or predicting which settings are sub-optimal.
According to the paper, these agents would bring 3 key advantages:
- Real-time collaboration: The agents work together in real time, sharing insights and making adjustments as the print progresses. Errors are caught and corrected immediately, reducing waste and improving quality.
- Learning and adapting: The framework isn’t just “set and forget.” It optimizes each layer and improves its performance over time — like having a team of experts continuously fine-tuning the process.
- Autonomy and scalability: Because it doesn’t rely on training data, the system can scale easily across a variety of printer setups.
In practice, how do AM companies adapt their software architecture to enable or support agent-like workflows?
In practice, we believe that software and SaaS providers will be among the most impacted by the rise of AI agents — and will need to adapt quickly to maintain their value across the manufacturing chain. A key step will be rethinking their software architecture to enable and support agent-like workflows.
So, how is it done at Authentise?
“We ensure that our architecture is flexible enough to use multiple different models (depending on the company or use case). We’re also paying even closer attention to weave together the digital thread in order to create sufficient data for these models to use,” Wegner told 3D ADEPT Media.
To mention a project where they have been working on, he adds: “We generated $8.1m savings for Boeing in the first year of operating ThreadsDoc, an AI-powered tool that captures additive engineering data and automatically generates templatised reports. In the case of Boeing, this was Technical Data Packages necessary for flight approvals, for which the engineering work was already done but which had to be written up. ThreadsDoc allowed highly skilled engineers to save 100’s of hours by getting a draft ready for their review.
“We have always pursued a mission of connecting “contextual data” because additive, as a high mix/low volume manufacturing process, does not have sufficient data for us to learn from traditional Industry 4.0 data (single sensor data) alone. So the emphasis has to be not only to integrate additive machines, but post-processors on the backend, and simulation engines as well as other design tools on the frontend,” Wegner points out, commenting on their integration strategy with other software or hardware in the AM value chain.
And at Synera?
“Our architecture, from the very start — with the visual construction of automated processes — already lends itself to building workflows that can use agents. So for us, the only real addition we needed to make was the integration of large language models, essentially creating the connection to the infrastructure required for agents,” Sartorelli said, speaking of Synera’s case.
“I think where we’ve seen AI agents deployed in the engineering context — and you can imagine this for additive manufacturing — there are a lot of different steps in the process, a lot of different tasks, and many individuals involved. What we’re seeing now is agents performing costing operations or leveraging nesting algorithms to find the most optimal nesting, orientation, and similar parameters. The nice thing about agents is that they can work somewhat autonomously to handle the mundane tasks that engineers are often forced to do manually by clicking around in their software. Agents remove that burden and instead operate based on the intent of the engineer or designer. So, if you don’t have a connection to the engineering tools you’re using for 3D printing, then agents can’t really help you. That’s why, from the very start, we’ve focused on integrations with partners like EOS, Materialise, Cognitive Design Systems, and Hexagon — to bring all of these tools together. Now, we see an opportunity with agents to leverage other integration strategies, such as the new MPP standard, to make these connections easier than ever before,” he adds.
The company is currently working with Materialise so that its users can access Magics SDK and deploy additive manufacturing agents that handle design-to-print tasks autonomously. Using a visual editor, engineers can automate at multiple levels of complexity – from simple tasks to complex AM end-to-end workflows thanks to fully-fledged multi-agent systems that seamlessly integrate with tools like Materialise’s Magics SDK. This will enable users to create end-to-end automation workflows for additive manufacturing, significantly reducing build failures and ensuring models are properly prepared before printing.
Current limitations of AI Agents in AM
Our assessment reveals that the limitations of AI agents in the AM chain could be similar to those faced by digital twin technologies: as there needs to be real cohesion and dynamics between the digital and physical environments. This means that:
AI agents should be powered by models that must understand and respond to both cyber and physical environments fluently. And the problem is, physical environments are variable and depend on a lot of factors, including changes in vibration or temperature, material quality, or human knowledge.
That’s probably what Authentise’s CEO summarizes as “accuracy”: “Accuracy is another key one, besides the one that you mention. This means that there are certain use cases (like full-text to CAD) that are unlikely to be particularly successful in the short term.
Externally, standards requiring a human in the loop will also slow the deployment of AI down until it can be proven to deliver a repetitive process. This will take time.”
“Now it’s really a matter of figuring out what the right use cases and applications for AI agents are. I don’t think we’ll see AI agents completely eliminating the need for people in the loop anytime soon. But over the next months or years, I do expect some routine tasks to be eliminated thanks to AI agents. If we skip the design phase — which, in my view, still requires the eyes of a design engineer — then yes, agents will start to cover the rest of the end-to-end workflow. However, there will always be a need for a human in the loop at certain stages. As engineers, designers, and scientists, we also have to validate the work of others — whether those others are people or AI agents,” Synera’s expert concludes.
It’s still too early to draw definitive conclusions about the integration of AI agents into the AM value chain. While we will keep monitoring this evolving trend, one thing is clear: those who want to stay at the forefront of the industry will need to cultivate an organizational culture that embraces and co-creates with AI.
*This dossier has first been published in the 2025 July/August edition of 3D ADEPT Mag. You may access the entire mag here.
