Design World moderated this year’s Future of Engineering Summit on March 25, 2026. This virtual event is a collaboration among leaders and engineers transforming technology, engineering workflows, and how teams generate value.
To view sessions from the 2026 Future of Engineering Summit on demand, visit future-of-engineering-summit.com/recap/spring-2026.
Why do agentic AI implementations fail? Are they too costly? Is the technology still too immature? Gartner predicts that more than 40% of agentic AI projects will be canceled by the end of 2027, citing escalating costs, unclear business value, and inadequate risk controls.
Agentic AI — autonomous systems capable of making decisions and executing complex tasks — is no longer the stuff of chatbots. It’s being integrated into simulation, design, and manufacturing systems to drive measurable industrial impact. But moving from proof of concept to production remains the central challenge, and it was the central theme of the 2026 Future of Engineering Summit, held March 25.
“This is not because the technology doesn’t work,” said Ryan Qi, principal worldwide BD/GTM leader at AWS. “It’s because we see a lot of organizations jump into multi-agent systems without the right foundations. The companies that succeed will be the ones who get the architecture, the security, the governance right at the start.”
The summit focused heavily on the transition from experimental AI to agentic AI and its role in removing the “human bottleneck” in industrial workflows.
“We believe generative AI is how organizations can fully recognize the full promise of large language models,” Qi said. “But generative AI alone won’t do it. Agentic AI will be the one unlocking it.”
Start with high-value, low-complexity applications
The summit opened with a session on “Turning AI into Measurable Industrial Impact,” led by Luca Zampieri, engineering director U.S. at Neural Concept, and João Moura, senior application engineering manager at Neural Concept. They believe that the era of isolated AI experimentation is over. For AI to deliver a competitive advantage, it must be embedded directly into engineering workflows as a structural capability — moving toward AI-native systems that can handle multi-physics surrogate modeling and real-time generative design.
“What does ‘efficiency’ mean in the age of AI agents? We’re talking about return on investments. It’s not only about having good results,” Zampieri said. “An agent that is made by a large language model — a set of tools that it needs to call, and a loop that iterates across these tools — can request much more tool-calling than what a human in an engineering discipline might request. And if the bottleneck in your industry is the simulation, for example, that can become very expensive.”
After walking through technical case studies on what Neural Concept calls engineering intelligence, Zampieri passed the mic to Moura for a deep dive into enterprise impact. Moura said engineering organizations care most about three core KPIs: shortening time to market, increasing engineering productivity, and innovating better products that cost less to produce.
“The reality check, however, is that the majority of AI initiatives do not reach production, and only a small fraction delivers measurable ROI at scale,” Moura said. “Innovative companies put intelligence at the center, and they leverage AI to accelerate their engineering processes, move fast, and achieve success at scale.”
His advice is to shift from a proof-of-concept mentality to a proof-of-value mindset and start with high-value, low-complexity applications to score quick wins.
“The question is no longer if AI will transform engineering. It is really, ‘Who will manage to scale it first?’” Moura said.
Move humans above the loop
Qi and Dr. Marc-Florian Uth, strategic partnerships lead at Synera, discussed how AI agents are transforming product development from simulation to manufacturing by automating repetitive tasks and optimizing workflows.
“Human engineers will be shifting from human-in-the-loop, where you have to manually trigger everything, to human-on-the-loop,” Qi said.
The shift is meant to free engineers from technical drudgery so they can focus on high-level strategy. But paradoxically, success can cause paralysis.
“We can replicate full departments or full process flows with several experts involved,” Uth said. “We have really good impact, but then it gets stuck. And the reason is that we are not prepared for such a big impact by just a single agentic workflow automation.”
Teams often lack the stakeholders and governance needed to scale pilots, Uth noted, so they remain stuck. Roles and responsibilities have to be clearly defined, a roadmap must be laid out, and teams need to recognize that agentic AI will fundamentally change their organizations.
“One exemplary change will be that in the past, we ran iterations between different experts who brought something to a decision committee,” Uth said. “In the future, we will go into a world where we have the human above the loop controlling one or several multi-agent systems, and is still responsible for the result, but has strong support and brings this to the decision committee.”
Bruno Finco, co-founder and CTO of MOVEdot, framed the underlying goal directly: “The goal of the AI layer in engineering is removing the human bottleneck, allowing innovation to move at the speed of computation rather than manual intervention.”
As an engineer and manager, Finco has seen firsthand the dramatic shift that occurs when an AI layer is added to data analysis.
“When we build the AI layer, we create this interface between engineers, managers, and all of the data that is available,” Finco said. “We end up with this unified evidence base that is not the job of the engineer, of the human, to navigate. No, the job of the engineer is to think of the methodology, bring expertise, have the right questions, and then have the AI answer those questions.”
Find puzzle pieces AI can solve
Jakob Lohse, senior product manager at Autodesk, recommends first identifying the puzzle pieces — the functionalities — needed to enable AI agents, and then exploring what AI can do for the team. He walked attendees through an automotive example at the interface of engineering and design.
“We always had this vision that people from different departments could meet in one room and work interactively on designing a car, and do everything within the meeting without having to postpone decisions because they lack information,” Lohse said. “One way we could overcome those time-consuming iterative cycles is obviously using AI models to store existing company knowledge.”
Lohse noted that ten years ago, interactive design was mostly manual. Then, deep learning let engineers design shapes and surfaces and, with a single click, get performance predictions to evaluate. But engineers would hit a wall when they were unhappy with the results, speed, or efficiency.
“This is the point where we want to use AI to generate surfaces,” Lohse said. “This is what we call a recommendation system, which basically couples predictive AI with generative AI, where you can create unlimited new geometry variations and evaluate the performance of each.”
The human engineer still evaluates the results and decides whether to trust them or rerun the optimization. Even with reruns, the cycle is faster, and multiple teams can collaborate in real time.
As Finco put it, AI agents “are working with you. They’re not just there as an assistant. They’re becoming a colleague. You interact, you debate, you question, you ask for, you engage — and with this, you are building knowledge.”
Finco described exponential growth through this collaboration, where organizations can deploy hundreds, even thousands of agents, working together at scale — not just collecting information, but taking action and improving.
Deploy strategically, and get a champion
To close this year’s Future of Engineering Summit, Uth, Lohse, Finco, and Nina Korshunova, head of AI engineering at appliedAI, gathered in a panel on where agentic AI can create real, measurable impact today.
“There has to be a strategic intent,” Korshunova said. “There has to be C-level sponsorship and an understanding that this is not just a project — it’s a whole rethinking of all the processes, organization, culture, tools, architecture. It’s a holistic approach and understanding.”
Though the need for a clear strategy resonates, teams often fear they are not ready or do not have enough structure in place to scale.
“It’s very important if you really want to scale this, but maybe it’s not so important to begin with it,” Uth said. “You won’t really deploy agentic AI, then it will be everywhere in your organization. You need to go step by step. You will start with one agent, and then one multi-agent system does one job, and you take it from there.”
In other words, build the structure as the agent system grows. The goal is not to deploy everything quickly, but to deploy in a controlled, strategic manner. A champion to lead the work also helps.
“My biggest success with customers is when they have someone responsible for that implementation — not only on the provider side, but internally,” Finco said. “That is ultra important because then you have someone who understands the system, who understands the engineers that should be helping them implement and validate.”
The panel agreed on the importance of strategically selecting the right starting point and expanding implementation as it makes sense for the organization.
“At the engineering level, the first challenge is always to enable those features that are controlled by an agent to train your own AI models based on your own data, and that you have sufficient datasets available for this in a structured way,” Lohse said. “Second, a very important puzzle piece is that you have sufficient API functionalities that can be orchestrated by an agent.”
That requires a holistic look at an organization’s current systems and the best path to making them agentic-friendly.
“You need to bring your people along because this requires cultural change, and cultural change starts at the top,” Korshunova said. “You need to go one step above and unblock the top-level process for your organization.”
The main takeaway from the 2026 Future of Engineering Summit is that the 40% failure rate that Gartner predicts is not a verdict on the technology, but on how organizations approach it. Strategic intent, clear ownership, the right starting point, and a willingness to build structure as you scale — these are what separate the agentic AI projects that reach production from the ones that get canceled.
To view sessions from the 2026 Future of Engineering Summit on demand, visit future-of-engineering-summit.com/recap/spring-2026.
Filed Under: AI Engineering Collective, AI • machine learning, ENGINEERING SOFTWARE
Tagged With: autodesk, AWS, neural concept, synera