One problem with the way we talk about AI and professional work is that we focus too much on what autonomous systems can achieve, and not enough on what we are willing to trust them with. Just because an AI model can generate a plausible answer, or an agent can complete a workflow, does not mean we have resolved the harder question of responsibility. Who decides when the output is good enough? Who understands the trade-offs? Who is accountable when the system fails? Engineering brings that tension into sharp relief. AI may generate the design, but a human still has to sign.
I have recently completed a major research project for the ACEC Research Institute, as part of its Firm of the Future series, exploring how AI, talent shortages, data, and digital transformation are redefining the engineering firm itself. The report (which you can read here) was grounded in detailed interviews conducted across engineering firms, technology providers, and the broader AEC ecosystem, including leaders from Autodesk, NVIDIA, WSP, Mott MacDonald, Bentley Systems, Esri, BST Global, Moffatt & Nichol, and others. We combined those conversations with prior ACEC research into AI adoption and the future engineering workforce, as well as Tomorrow’s own proprietary research into digital labor and the changing design of firms. The result was not a technology forecast in the usual sense. It was a bottom-up narrative account of how leaders inside a conservative, safety-critical, highly traditional industry are beginning to think differently about work, value, talent, and the future operating model of the firm.
The bottom-line finding was interesting. AI is already changing engineering work, but not primarily by replacing engineers. The more immediate effect is that it changes the ambition of what can be achieved. Engineers are beginning to move from manually producing a small number of design options to orchestrating systems that can generate and test thousands. They are spending less time on repetitive documentation, drawing production, and knowledge retrieval, and more time framing problems, evaluating trade-offs, validating outputs, and advising clients. In the most sophisticated firms, AI is not just making the old work faster. It is expanding the solution space. It allows engineering teams to ask a different class of question: not merely, “Can we design this?” but “Of all the technically viable ways this could be designed, which one best balances cost, resilience, sustainability, safety, speed, and long-term performance?”
The engineering industry is already under enormous pressure. Demand for infrastructure is accelerating, driven by aging systems, climate adaptation, electrification, urbanization, and the rise of digital infrastructure. The AI economy itself depends on a massive physical buildout: data centers, power generation, transmission networks, cooling systems, water infrastructure, and the industrial supply chains needed to support them. The bottleneck to AI’s future may not only be faster chips, better networking, or more efficient models. It may be the human and institutional capacity to design, permit, build, and operate the infrastructure that makes AI possible.
That is why the tightening supply of engineering talent could not be happening at a worse moment. Prior ACEC research highlighted the imbalance clearly: in the U.S., roughly 184,000 engineers left the workforce in a single year while only around 166,000 new engineers were available to replace them, creating a net shortfall of about 18,000 professionals. Globally, some interviewees in our research pointed to the possibility of a much larger gap in the workforce required to deliver critical infrastructure by 2035. Just as AI is increasing the demand for physical infrastructure, the industry responsible for designing and delivering that infrastructure is running short of the people it needs. In that context, AI is not simply an optional productivity tool. It is becoming a response to a structural capacity problem.
Automation alone will not solve the capacity constraint, because engineering is not only about producing more work. It is also a question of accountability. Even in a world where AI agents can generate designs, run simulations, identify clashes, check compliance, draft documentation, and manage workflows, the need for a human engineer to take responsibility is unlikely to disappear. Engineering is not like generating marketing copy or summarizing a meeting. A design decision can shape public safety, asset performance, environmental outcomes, and capital investment for generations. Clients are not merely buying calculations. They are buying trust. Regulators, insurers, and courts will still want to know who reviewed the output, who understood the assumptions, who accepted the risk, and who signed the drawing. The machine may generate more possibilities than any human could produce, but a person still has to decide which possibility is appropriate.
As creation becomes more abundant, validation becomes the scarce and strategic act. In a traditional workflow, design generation was the bottleneck. Engineers could only explore as many options as time, budget, and human capacity allowed. In an AI-enabled workflow, that constraint begins to move. The problem is no longer whether the system can produce enough possibilities, but whether the firm can test, verify, and stand behind them. As Julien Moutte, CTO Bentley Systems explained to us, the faster designs can be generated, the more critical it becomes to ensure they are right. “You need to test everything,” he said. “That’s what gives engineers the confidence to sign the design.”
Greater AI autonomy does not diminish the profession. It elevates it. But it also raises a difficult question for firms: how do you train engineers to develop judgment if AI automates many of the early-career tasks through which judgment used to be built? If young engineers no longer spend years doing calculations, reviewing drawings, and working through the details of delivery, firms will need new forms of apprenticeship, simulation, mentoring, and deliberate practice to build the intuition that accountability requires.
The firms that thrive in this environment will not simply be those that buy the best AI tools. They will be those that redesign themselves around a world of abundant intelligence. NVIDIA’s perspective was especially revealing here. From its vantage point at the foundation of the AI infrastructure stack, the future engineering firm begins to look less like a conventional professional services business and more like a hybrid human-AI organization.
When we spoke to Sean Young, Director of AECO, geospatial, and AI solutions at NVIDIA, he described a world in which each engineer may be supported by dozens of specialized AI agents. One agent might generate geometry. Another might run structural analysis. Another might check code compliance. Another might estimate cost or schedule impact. The engineer becomes the supervisor of a digital workforce, not merely the user of a software application.
That idea has profound implications for the structure of the firm. A company with 10,000 employees could eventually be managing hundreds of thousands of digital agents. Today, IT departments manage devices, networks, software licenses, and cybersecurity. In the future, they may manage digital labor: agents that need to be deployed, governed, monitored, evaluated, and improved. The organizational question shifts from “Which tools do our people use?” to “How do humans and machines collaborate as a single productive system?” In that world, competitive advantage depends on proprietary data, simulation capability, AI governance, and the ability to encode institutional knowledge into systems that amplify human expertise. The firm of the future is not defined by headcount alone. It is defined by how effectively it organizes intelligence.
Autodesk’s future roadmap points to an equally important reversal in the work process itself. For decades, the AEC industry has moved through successive waves of digitization: from paper to CAD (Computer-aided design), from CAD to BIM (Building Information Modeling), and now from model-based design toward outcome-based BIM. Each transition changed the interface between professionals and their tools. Paper made design physical. CAD made it digital. BIM made it coordinated and data-rich. But AI changes the direction of the process. Traditionally, engineers and architects constructed a model, then tested it against requirements. In Autodesk’s emerging vision, the professional starts by defining the desired outcome. For example, a hospital might need a certain number of rooms, a net-zero target, a fixed budget, a construction timeline, and particular performance requirements. The system then generates and simulates a vast number of possible designs that meet those constraints.
Outcome-based BIM is a fundamentally different way of working. The professional is no longer drawing a solution one element at a time. They are shaping the space of possible solutions. Nicolas Mangon, Vice President AEC Strategy at Autodesk described this to us as a shift in which “the desired outcomes are the input to the process.” The software becomes less a drawing tool and more an exploration engine. For standardized asset classes such as warehouses, housing, and data centers, this could push the industry toward industrialized, repeatable, highly automated design and delivery. For complex, one-off projects, it will not remove human expertise, but it will concentrate it where it matters most: judgment, trade-offs, client context, and accountability.
The same shift is also pushing firms beyond the boundaries of project delivery. A digital model created during design no longer has to disappear once construction is complete. It can become an operational twin, connected to sensors, performance data, maintenance systems, and AI analytics. That changes the role of the firm. Instead of handing over an asset and walking away, engineering firms can remain connected to the infrastructure they helped create, helping clients understand how it performs over decades. In that world, the most valuable output of a project may not be the drawing or the model, but the intelligence that accumulates around the asset over time.
Ultimately, this is a discussion that is not only relevant to people who work in the architecture, engineering and construction industry. Many other professionals in law, consulting, accounting, finance, and technology all face versions of the same question. What happens when digital systems can perform more of the work, but humans remain accountable for the consequences? What happens when the old apprenticeship pathways are disrupted? What happens when value shifts from producing outputs to selecting, validating, and standing behind outcomes? And what happens when firms built around labor scarcity suddenly have access to scalable intelligence?
The deeper lesson is that AI will change where value lives. In engineering, the greatest risk is that the intelligence embedded in projects, data, workflows, client relationships, and hard-won judgment migrates into the platforms around the firm. The same challenge now confronts every expert organization. The firms that win will be those that redesign themselves around what they know, how they decide, and why clients still trust them when machines can generate credible answers.