Summary
The adoption of artificial intelligence (AI) is driving a profound transformation in architecture, engineering and construction (AEC). In particular, the coordination of mechanical, electrical and sanitary installations (MEP) is benefiting from advanced algorithms inspired by challenges such as the Japanese Maze Race, allowing the optimal automatic tracing of routes. This article integrates a specific case of MEP coordination with AI - based on a technical article released in 2023 - with a broader overview of how AI is revolutionizing the AEC industry. Emerging tools such as Spacio, Augmenta, and MagiCAD are explored, analysing their integration with the BIM methodology and the Virtual Design and Construction (VDC) processes both in the early design phases and during on-site execution.
Introduction
In recent decades, the AEC industry has experienced a significant advance in digitization through the Building Information Modelling (BIM) methodology and Virtual Construction (VDC) practices. However, important challenges persisted, particularly in the efficient coordination of MEP facilities within complex BIM models. The superposition of ducts, pipes, electrical conduits and other elements in small spaces usually generates conflicts (clashes) that, if not resolved at the design stage, cause expensive reprocessing and delays during construction. Traditionally, MEP coordination required long manual review sessions and iterative adjustments between disciplines. Faced with these challenges, artificial intelligence has emerged as a promising tool to automate and optimize coordination, reducing errors and accelerating workflows.
An inspiring milestone in the development of automatic routing algorithms comes from an unexpected area: Micromouse's competencies (popularly known as the Maze Race). Since the late 1970s, these contests have challenged small autonomous robots to find the fastest route through a maze. The robot must explore the environment without human assistance, mapping walls and corridors and then, after several explorations, execute the optimal route in record time. In fact, the current world record in this test is just 3,921 seconds, achieved by a robot that applied efficient path-finding algorithms. The algorithms used - variants of methods such as flood-fill, Dijkstra or A* - allow you to evaluate multiple routes in parallel, assigning scores or "costs" to each cell or section of the maze, to finally choose the route with the lowest cost.
This algorithmic logic has found a remarkable parallelism in AI-assisted MEP coordination. In a facility project, the structure of the building and the possible paths for pipes or ducts resemble a three-dimensional maze. AI can "explore" different routes to connect, for example, an HVAC equipment with a network of ducts or an electrical box with its loads, avoiding obstacles such as beams or walls. Following a logic similar to that of the Micromouse, some algorithms perform multiple passes: the first exploratory to generate a heat map of possible paths with their costs (distance, changes of direction, congestion, etc.), and a last pass to choose the optimal pathlinkedin.com. This avoids dead-end paths or unnecessary loops, in the same way that the robot mouse learns not to repeat routes already visited. The incorporation of vertical movements (ups and downs between levels) adds complexity, but the principle remains: AI seeks the most efficient route in a complex three-dimensional space.
In this article, we will analyse how these techniques have been implemented in real AI-assisted MEP design tools, linking the specific case described by Domínguez Rogers (2023) on MEP coordination with the broadest advances in the AEC industry. Platforms such as Augmenta, which applies generative AI to the design of electrical paths, MagiCAD, which integrates machine learning algorithms in MEP modelling, and early generative design solutions such as Spacio will be discussed. In addition, we will explore the integration of these innovations with BIM and VDC: on the one hand, in the early phases of design (conceptual and detailed) and, on the other, during on-site execution through 4D/5D synchronization, computational vision and intelligent construction management.
MEP coordination with AI: from labyrinth to work
Traditional MEP coordination usually requires manually iterating the location of pipes, electrical trays, air conditioning ducts and other components to resolve interference. This process can be slow and prone to human error. AI offers a new automated approach: algorithms capable of tracing optimal routes for facilities within the BIM model, meeting criteria of spatial efficiency, material minimization and regulatory compliance.
Fig 1 MEP HVAC development with VectorWorks and optimized layout with Augmenta
An illustrative antecedent of how AI addresses the problem is the analogy with the Maze Race described above. Domínguez Rogers (2023) highlights that MEP coordination algorithms conceptually originate in competencies such as Micromouse. In these skills, the algorithm assigns a value to each box in the maze during the initial explorations, indicating the "distance" to the objective; then, in the final execution, the robot follows the sequence of cells of lower accumulated cost. Applied to a building, this approach involves discretizing the space (for example, in a three-dimensional mesh or graph of interconnected nodes by possible paths) and using an AI to learn which routes offer the lowest cost in terms of length, number of elbows, crossings with other facilities, etc. A technical article by Mathieu Josserand (2021) details how a Deep Q-Learning agent trained in Revit models can automate the layout of MEP accessories, analysing the geometry of the plant as a matrix and learning the most efficient paths through trial and error guided by rewards. In essence, AI "learns" to design installations by repeating route simulations just as a robotic mouse would learn a maze.
Augmenta case: automatic electric tracing
One of the pioneering tools to bring this logic to the real world is the Augmenta platform. As a result, the tool manages in a short time to find the "shortest and cheapest route to build" for electrical pipes.
Fig 2 Modeling optimized by Augmenta
A case study published in 2024 showed the tangible impact of Augmenta on real projects. Miller Electric Company, an electrical engineering firm in the United States. This saving is equivalent, according to the company, to the working capacity of 2-3 additional BIM modelers without increasing the equipment. In addition, the designs produced by the AI were directly usable for manufacturing, incorporating a high level of development (LOD) details, including the estimation of material and labour. This suggests that automated MEP coordination not only streamlines the design but also produces constructive and shock-free BIM models ready for the next stage of the process (prefabrication and assembly). However, it is also reported that, being a technology in early stages, the system can present failures or suboptimal solutions in some cases. Augmenta has recognised this situation, inviting companies to contribute real project cases to their pilot phase to train and strengthen AI with more data. This reflects a crucial aspect of modern AI solutions: their continuous improvement depends on learning from large data sets and feedback from real users.
MagiCAD case: machine learning integrated with BIM
Another outstanding initiative is that of MagiCAD, a widely used MEP design software that has begun to incorporate AI algorithms in its latest versions. MagiCAD has an extensive BIM library of more than a million objects from real manufacturers (HLC equipment, pipes, luminaires, etc.), and its AI approach focusses on accelerating and facilitating repetitive MEP modelling tasks. According to information disclosed by its developers, MagiCAD is implementing machine learning (automatic learning) and deep learning techniques to automate “complicated, slow and inconvenient" MEP design jobs, making them "much easier, cheaper, more manageable and faster." The stated goal is to make the MEP design faster, more efficient and more accurate through intelligent automation. In practice, this could translate into functionalities such as self-routing of pipes between two given points avoiding collisions, suggestion of optimal locations for diffusers or electrical panels based on previous experiences, or automated sizing of networks according to calculated loads.
Fig 3 MagicCAD in AutoCAD
MagiCAD is adopting a learning approach based on data provided by users themselves. The MEP projects uploaded or synchronized on its cloud platform serve as training material for the AI to "learn what is right, what is not, and how to make it more efficient." In the words of the MagiCAD team: "we are building new tools for designers who use state-of-the-art methods to improve MEP design and modelling step by step." This suggests that MagiCAD could, with enough accumulated information, transform into a kind of intelligent assistant within Revit or other platforms, which recommends optimal solutions to the modeler or even proposes complete sections of facilities. It is worth highlighting the analogy used by Domínguez Rogers: today, the AI in BIM design is like "a 2-year-old boy... who is still learning to walk." That is, he makes mistakes and does not yet have the precision of a human expert, but he is in an intense phase of learning. With time and training, it is expected that these functionalities will become more reliable and robust.
Early generative design: the contribution of Spacio and others
If Augmenta and MagiCAD address the optimization of detailed facility design, other AI tools focus on earlier stages of the project process. Spacio is an example of a generative design platform that combines AI with parametric design to assist architects in the conceptual phase. Unlike traditional BIM tools (e.g., Revit) that prioritize precision and detail, Spacio seeks agility and rapid exploration in the preliminary design. It allows you to generate in minutes multiple architectural preliminary project options complying with regulations (building codes) and objectives defined by the user. Its intuitive and collaborative web interface reduces the typical learning curve of complex parametric environments, democratizing its users. Although Spacio is mainly oriented to architects, its relevance in this context lies in the fact that it produces "BIM-ready" models from the beginning, that is, volumetrics and spatial distributions that already contain structured information and can serve as the basis for subsequent engineering developments. In this way, an integrated workflow could start from an AI-generated design (Space or another similar), then move to an AI-assisted MEP detailing (Augmenta/MagiCAD), maintaining the traceability of information in a unified BIM environment.
Fig 4 Generative Design of Urban Complex with Space AI
Generative design is also venturing into MEP and structural engineering through platforms such as Hypar. Hypar offers an open environment where personalized generative workflows can be built for different project requirements. Recently, Hypar has applied AI to automatically generate multiple MEP design options optimized in cost, performance, and sustainability. A notable feature is the early integration of operation and maintenance considerations: Hypar allows the incorporation of virtual IoT sensors and predictive maintenance schemes from the design phase, so that the designed facilities not only work efficiently when inaugurating the building, but are easy to maintain in the long term. This reflects a convergence between generative design and intelligent building operation, an area known as Facility Management 4.0. The platform even conceives that contractors can use these models to generate new revenue by offering AI-supported maintenance services (performance monitoring, replacement alerts, etc.).
AI, BIM, and VDC synergy
The integration of AI in the AEC industry does not occur in a vacuum, but supported on the already established bases of BIM and VDC. BIM provides the information-rich digital model on which AI tools can operate, either to read geometric conditions (available spaces, interferences) or write optimized design solutions. VDC, for its part, extends the use of the digital model to the planning of construction, logistics, costs, and operations. Next, we examine how AI solutions are embedded throughout the project life cycle, from design to on-site execution, empowering existing processes.
Fig 5 REVIZTO in action
BIM design and coordination phase
In the design phase, AI acts as an assistant and automator within the BIM ecosystem. We have already seen how generative design tools accelerate the conceptual phase and how automatic MEP coordination solutions solve complex details. A critical aspect is to ensure that these proposals comply with regulations and are free of coordination conflicts. This is why AI-powered crash detection comes into play. Model validation applications such as Solibri have incorporated intelligent algorithms to review BIM models for collisions between systems and to verify automated regulatory compliance. Solibri uses predefined and customizable rules (for example, verify minimum safety distances or dimensions of corridors according to code) and scans the model by identifying both geometric and business rule conflicts. AI allows you to process complex models with thousands of elements, detecting subtle scenarios that could escape the human eye, while generating detailed and actionable reports for the design team. This not only covers physical shocks, but also regulatory violations before arriving at work, avoiding delays or costly modifications for non-compliance.
Another front is the integration with the manufacturing detail. Tools such as Trimble SysQue, although they are not AI in themselves, complement the BIM->Manufacturing flow, ensuring that the coordinated model includes real and detailed components (nuts, flanges, supports, etc.) from specific manufacturers. Once the AI has arranged, for example, the optimal trajectory of a pipe, SysQue allows you to replace that generic element with the exact commercial part (90° elbow of a certain radius, for example) that will be used on site. This produces a model to the LOD 400 or higher, ready to generate cutting sheets, purchase orders or assembly guides. The automation of coordination must therefore be accompanied by this translation into the physical world. In this sense, AI coordination solutions are beginning to feed on real catalogues: for example, Augmenta incorporates material specifications and local electrical regulations when generating its designs, and even optimises according to the cost of materials and labour on each route. The result is a BIM design "with awareness" of costs and feasibility, which responds quickly to changes (if a major modification occurs in the architecture, the AI can re-calculate a new solution in minutes, maintaining coordination).
4D/5D planning and work management
During planning and execution, AI continues to add value to the digital model in what is known as 4D (time) and 5D (cost) dimensions. A recent advance is the automation of the link between the 3D model and the work schedule (4D). VDC software such as Fuzor 2024 introduced AI-driven workflows that replace the manual task of linking each BIM element with program activities. Fuzor's "Expert AI" engine allows, after loading the architectural/engineering model and time planning (for example, from MS Project or Primavera P6), the software to automatically generate the animated constructive sequence, assigning elements to each phase in a matter of seconds. This frees planners from tedious and omission-prone work hours. In addition, Fuzor incorporates a concept of micro schedule (microprogramming) where the AI identifies detailed tasks that do not appear in the high-level schedule but that are necessary on site (assembly of scaffolding, movements of equipment, etc.), inserting them in the 4D simulation to detect potential conflicts ahead of time. For example, AI can suggest the path of a mobile crane inside the site and predict whether it will interfere with the installation of MEP pipes at a certain level, adjusting sequences to avoid temporary collisions. These capabilities illustrate how AI and VDC complement each other to achieve finer, more proactive planning, reducing on-site improvisations.
As for 5D (costs), AI is focusing on both the estimation and the control of expenses. In the bidding phase, models trained with historical data can predict costs more accurately and propose optimizations. For example, the Norwegian company AF Gruppen integrated AI into its planning flows for a skyscraper in Oslo (~$560 million); the AI analyzed multiple combinations of constructive sequences, crews, and supplies, generating seven schedule alternatives that exceeded the initial plan in efficiency. The company selected the best option and managed to reduce the total cost of the project by 15% compared to the original planning. This case, related to the ALICE Technologies platform, shows that AI can find solutions that shorten the duration of work (in that case, a decrease of 18% in time was reported) and consequently lower indirect costs, also optimizing the use of resources.
During construction, maintaining control of what is planned versus what is executed is key to avoiding cost overruns. Here come the computational vision and IoT applications for real-time monitoring. Tools such as OpenSpace and Buildots use 360° cameras and visual recognition algorithms to create updated digital twins of the work. An engineer can virtually tour the project through immersive photographs taken daily, which the AI automatically positions in the plans with accuracy. This allows deviations to be detected: for example, if a section of MEP pipeline was not installed where it should on a certain date, the system highlights it by comparing the real image with the planned model, alerting the management team. In addition, in November 2024 Buildots launched "Dot", a virtual construction assistant capable of answering questions from managers about the status of the project (for example: "What is the progress on floor 3?") Analyzing the captured data. These capabilities streamline decision-making and communication, since any deviation or problem is identified before it escalates into cost or delay.
The Internet of Things (IoT) combined with AI also optimizes on-site management. Sensors in machinery and wearable devices in workers collect data on productivity, location, and security in real time. AI can analyze, for example, if certain equipment is underutilized or stopped, reassigning it to necessary tasks (resource optimization), or predict maintenance to avoid breakdowns that stop the work. In the field of security, AI vision systems monitor the use of personal protective equipment and the presence of risks; if an operator enters a dangerous area without a helmet, the AI can issue an immediate alert. These interventions not only protect lives but also reduce costs associated with accidents or interruptions.
In short, AI - supported by the basis of BIM models and VDC practices - is enabling a continuous improvement cycle: from an optimized design that translates into a more fluid construction, to an intelligently monitored construction that feeds back data to perfect future designs. The convergence of these technologies augurs works with less improvisation, less reprocessing, and greater predictability in terms and costs.
Conclusions
Artificial intelligence is changing the way we conceive, design, and construct our buildings and infrastructure. In the specific field of MEP coordination, AI has gone from being a theoretical promise to demonstrating in practice that it can solve problems in minutes that traditionally took weeks, even inspired by robotics solutions and labyrinth algorithms developed decades. Integrated with BIM and VDC methodologies, AI acts as a catalyst that accelerates workflows and reduces uncertainties throughout the project: from optimizing a preliminary project in its conceptual phase to monitoring work with a digital twin in real time. The cases reviewed between 2023 and 2025 show concrete benefits: time savings of 30-40%, reduction of direct costs of 10-15% in large projects, and qualitative improvements in coordination and security. At the same time, technology is still in accelerated evolution. As an expert pointed out, writing about the advances of AI in BIM today could require an update in just one year, given the speed with which new versions and tools emerge.
For AEC companies, especially in Latin America, getting on this swing of innovation represents a challenge and an opportunity. The challenge is to navigate the implementation with strategy: prepare your people, choose the right battles for AI and adapt processes. The opportunity is potentially huge in productivity and competitiveness. The region has the talent and creativity to adopt these technologies and, adapting them to their contexts, perhaps even develop its own solutions oriented to their markets. Leading regional countries in AI already show the way in investment in digital infrastructure and support policies. It is crucial that companies rely on that environment when it exists, and where it does not, that they themselves promote the conversation about digital transformation in the industry.
In conclusion, AI-assisted MEP coordination is a tangible example of the digital transformation of the AEC industry. Algorithms that learn to "run labyrinths" can now design our facilities; models that were previously static now dialogue with cameras and IoT to tell us how the work is progressing; virtual assistants begin to integrate into technical offices and construction fields. Far from displacing human work, these advances allow engineers, architects and builders to focus on tasks of greater added value: informed decision-making, innovation in design, and the creative resolution of non-trivial problems. AI is gradually imposing itself as a strategic ally – a new team member – that brings speed of calculation, analysis of large volumes of data and lessons learned from countless projects. Those organizations that embrace this ally with vision and preparation will be better positioned to build more, better and with less waste in the years to come. Like any disruptive change, there will be a learning curve, but the expected benefits in efficiency, cost and quality make the journey worthwhile. The race has begun; the invitation is open for the Latin American AEC industry to take the AI starter and join the global competition towards a smarter construction future.
References:
Domínguez Rogers, F. J. (2023) MEP Coordination and how AI is imposing itself. LinkedIn Pulse.
Josserand, M. (2021). How to automate MEP layout using Machine Learning. Medium.
Increases (2024). Case Study: How Miller Electric Reduced their Modelling Time by 40%. Augmenta Blog.
BIM Space (2025). MagiCAD, what is MagiCAD for Revit? [Web article].
NeevIQ (2025). Top 6 AI Tools for MEP in 2025: Transforming Workflows and Efficiency. NeevIQ Blog.
AEC Magazine (2023). Fuzor 2024 uses AI for "4D automation". Greg Corke
Numalis (2024). Driving Efficiency: How AI Streamlines Construction Costs. [Web article].
Open Space (2023). Product website and documentation. [Data on 360° monitoring].
Buildots (2024). Press Release: Buildots launches AI assistant for site managers. The Construction Index.
ECLAC - ECLAC (2024). Latin American Artificial Intelligence Index (ILIA) 2024 - Press Release.
Citations:
https://en.wikipedia.org/wiki/micromouse
https://futurearchi.blog/en/ai-parametric-design-space/
