SWAPP is revolutionizing the architecture ecosystem with its AI-powered platform which automates architectural design and planning processes.
By Martyn Day - AEC Magazine March / April 2024
In a world swimming in AI-related hype, several CAD software firms are making serious progress in the race to eliminate workflow bottlenecks using new technologies. Martyn Day looks at the field of drawing automation and identifies some of its frontrunners
By AEC Magazine
Throughout their history, desktop CAD systems have been sold on the promise of delivering one overriding benefit: improved productivity when compared to manual drawing.
That makes sense, since productivity improvements are always sought by firms looking to save money and to increase their competitive edge. In these highly digitised times, the deployment of technology has become a key differentiator in the AEC world.
While it took twenty years for most AEC firms to move from drawing boards to 2D CAD, many like to think that the adoption of 3D CAD and BIM has been faster. In fact, that move has also taken the best part of two decades. And many of those firms using BIM tools have still not progressed as far as they might, remaining resolutely 2D-centric. On the other hand, who can blame them, when the primary deliverable of most contracts – and the most frequently cited target in legal wrangles – are drawings?
BIM was sold to 2D CAD users as a way to model their buildings and get automatic sections and elevations as a base for drawings, with the added benefit of coordinated documentation updates when model designs change.
In fact, what has happened is that the number of drawings produced has mushroomed. Many BIM seats are not involved in design at all, but instead focus on documenting the relatively poor automated output of 3D models.
A cynic might applaud the CAD software firms for increasing the costs associated with a documentation seat and driving appetite for complex document management systems to handle greater volumes of drawings. After all, many firms take the output from BIM tools such as Revit and perform refinements to drawing output in AutoCAD – the king of the 2D drawing world. However, this undermines the benefit of keeping both models and drawings in the BIM environment, breaking the links needed to coordinate updates to sections and elevations.
As documentation mushrooms, and clients increasingly demand BIM deliverables (even if they only do so because that’s what their consultant advises), the result is a heftier workload that consumes available resources and delivers a serious hit to the bottom line.
Many design IT managers have told me in confidence that they wish that drawings would simply ‘go away’. They would prefer the model to be the deliverable from which all data is extracted by contractors and project managers.
But here, they are at odds with the industry’s foremost institutions, such as the Royal Institute of British Architects (RIBA) and the American Institute of Architects (AIA), not to mention complex legal frameworks that are still largely predicated on drawings.
Here to stay
Another question I hear regularly is, “Why can’t AEC be more like manufacturing?” The misconception here is that once manufacturing designs move into CAD, the importance of drawings is greatly diminished.
But it might shock readers to hear that many manufacturing sectors – and in particular, automotive and aerospace – are just as burdened by the need to produce manufacturing drawings to accompany their 3D models, typically for reasons relating to fabrication and legal compliance. And this is the case even where parts are 3D printed directly from a 3D CAD model.
With that in mind, it’s safe to say that drawings are simply not going to ‘go away’. They will continue to be essential deliverables, although uses may change and their relevance in some areas may decline.
This is not to say, however, that there is nothing to be done about this situation. In fact, modern mechanical CAD (MCAD) tools do a much better job of automatically creating drawings and reducing repetitive tasks than today’s BIM tools, even for very complex parts and assemblies. That said, even with this progress, it’s estimated that engineers spend between 25% and 40% of their time refining drawings.
This is frustrating, because in modelling a building, a car or an aircraft, most design and engineering decisions have already been made. That makes it hard to justify excessive amounts of time spent creating 2D drawings from a model, a process that adds no real value to a project.
And as complexity increases, the case for drawings starts to fall apart. Take, for example, Frank Gehry, who couldn’t get any of his buildings made until he began using Dassault Systèmes Catia and McNeel Rhino. The drawings were so hard to understand that fabricators would bump up costs in order to accommodate the impact of inevitable misunderstandings. When Gehry’s practice switched to modelling buildings in 3D CAD and started sending Catia models to fabricators, he saw a corresponding drop in quotes and bids were suddenly aligned to within 1% of each other. As a result, his buildings were viable – but Gehry still produces 2D section drawings in AutoCAD in order to detail more standard aspects of designs. In other words, models offer clarity in situations of complexity, but for most buildings, simple rectangles are hardly likely to bust budgets.
Auto-drawings on demand
When I first met Greg Schleusner, HOK’s director of design technology, at a Bricsys (BricsCAD) meeting in Stockholm, Sweden in 2019, we quickly got talking about automated drawings. He explained to me that this was something the industry desperately needed, but that nobody seemed interested in developing the technology.
The case he put forward was clear. At that time, HOK was spending around $18 million per year on creating and managing drawings and construction docs (inclusive of staff costs, software and so on). Schleusner reckoned that, if the firm could automate 60% of the work associated with construction documents, or maybe 50% of the work associated with drawings, it would cut those costs in half. Even partial automation had the potential to deliver significant cost benefits.
While researching the topic, Schleusner discovered a Korean open-source project conducted for a building authority that focused on taking in IFC data and creating floor plan drawings for fire code review. Labels, text, drawing grids and dimensions were all added by the programme when the automated 2D drawings were viewed. He was impressed by the quality of the results. Everything displayed was dynamically created on the fly. It didn’t exist in the dataset.
“I think, at least from my perspective, the problem is we import all this other stuff into Revit. That is so much of our design effort, just to waste time to make drawings that we must make. So not only are the drawings themselves costly, but we also have this interoperability problem that needs to be solved. We have to send model data between design tools to make drawings. An open auto-drawing tool would allow design teams to stay in Rhino,” said Schleusner.
“What do you need to make drawings? Do you need an accurate model? If we solve drawing automation, we are going to have to make better models. If we make better models – hey, we have better models! We are going to have to improve what we model to get a better drawing. Right now, we have so much drawing work, that we all end up ‘faking stuff’ in drawings, just to get them out the door. I actually think auto-drawings is a very complementary idea to making better models.”
Since 2019, Schleusner has been bending the ears of leaders at many of the biggest CAD vendors, trying to spark their interest in auto-drawings. Now, with the advent of automation and AI, those vendors are scrambling to identify where might be best to apply new technologies. Since drawings represent ‘low-hanging fruit’, interest in auto-drawings has exploded.
Most of the early auto-drawing solutions will begin by offering configuration-based output (drawing templates). But Schleusner is keen to see work begin further upstream, at the project level. This would involve understanding what drawing sets are required by project managers and project architects, not just automating ‘the drawing’ but also automating project document sets. This would require the equivalent of HTML and CSS for drawing content, settings and how it is all put together.
“If the industry can solve this auto-drawing problem as a service, it could shatter the market into one thousand solutions, which is perfect. I don’t think you’ll need monolithic applications, because drawings are external and it’s an assembly process.”
In passing, Schleusner has also lamented that PDF is the ultimate industry deliverable, as historic constraints of standard drawings sizes 1:100/200 and so on limit the digital canvas. He explains that one of the reasons the industry makes so many drawings is because of the physical limitations of displaying text legibly within paper documents. “Couldn’t we just send one bloody big PDF drawing, which contractors never print out? If you can’t see something, then zoom in! While certain trades really liked this idea, it’s true that some want to print things out, which is a limitation. But if we can’t get the digital delivery of drawings right, how are we going to ever do the digital delivery of models?”
Powered by AI
Eventually, I suspect AI will feature in most auto-drawing solutions. To start with, there will be procedural AI, providing configurations and so on. From there, we will see more companies attempting to use AI to learn drawing layouts and perform automation functions, such as auto dimension, auto label, auto title block, auto table, auto grid and so on.
To figure out where all this might lead, it’s useful to first look at Swapp. Based in Israel, Swapp’s initial application of AI to Revit promised us a miracle. What it proposed was to take users from 2D sketches to fully detailed BIM, with all the drawings involved, in the time it took them to go and have lunch.
Along the way, Swapp would learn from past projects and then automate Revit detail models – a proposition that favours standard, rectangular, predictable buildings, such as student blocks, hospitals and offices.
That was the message delivered by Swapp’s chief science officer Adi Shavit at AEC Magazine’s NXT BLD 2023. More recently, I caught up with Shavit to discuss auto-drawings. As he explained, the company has pivoted its software engineering focus to deliver the auto-drawing component as a bespoke solution, as well as expanding to Canada and the UK.
Shavit explained how, in talking with clients over the past year, the number one feature that they wanted to know more about was auto-drawings, and not so much about the magic at work in turning 2D sketches into fully detailed Revit models.
However, to understand how to make a drawing, Swapp needs the model data. “All of our processing is done on our system in the cloud. We have a plug-in for Revit, which does the export and the interoperability part, but nothing really happens on the desktop. It talks to our system and sends us the model, then we generate the construction documents from that data and bring it back into Revit drawings,” Shavit explained.
“We are working with architecture companies, and we essentially say, ‘You can think of Swapp like you outsource drawing production to a team in Belarus or in India, but our delivery times for drawings are constant.’ Meanwhile, our customers are saying that they are invoicing their customers sooner, by one month or two months. And for us, that’s the name of the game. We’re charging based on project size.”
When it comes to rules/template-based automation, Shavit told me that Swapp tackles the output standards problem from the opposite direction, as every office has its own standards, and even within firms, teams have different standards. Swapp treats the standardisation of output wider and at much higher resolution. There is no uniform Swapp standard. It is derived from past projects, to generate drawings that are compliant with the specific customer’s way of doing things. This can start with Swapp looking at its past projects (both the BIM and the drawings), or it can be trained on projects currently in the works.
Shavit readily admitted that the documentation that is generated is not 100% complete. There are always special cases, special fire requirements for certain parts of the building and other external requirements. Documents will still need to be put past a senior architect for liability issues. But he claimed the company is trying to perform between 80% and 95% of the most tedious, time-consuming work, enabling skilled staff to focus their efforts on other aspects of projects.
At its heart, Swapp is a consultative, bespoke auto-drawing solution that uses AI. Company executives are keen to state that it siloes every client’s data and AI into their own results. From this, it sounds as if Swapp has clients already using its technology, even though it’s hard to understand what they do from the company’s website.
It will be interesting to see how Swapp’s output quality, speed and ‘completeness’ compares to Autodesk’s own in-Revit auto-drawing features (more on this ahead) when they eventually ship.
Conclusion
Auto-drawings is not a matter of if, but when. It’s also a question of establishing which offerings are any good. There are so many firms now researching this area that auto-drawings are going to be available in pretty much every tool out there, both old and new. But there will doubtless be differences between them, in terms of capability, cost and how they are hosted.
Swapp is taking more of a bespoke approach and will create an auto-drawings system bespoke to your firm’s particular needs.
As far as timelines are to be considered, Swapp has demonstrated some capabilities from its cloud AI and has customers now. However, I think we are looking at two to three years before we have a choice of mature, tried-andtested auto-drawings technologies on the market, either in your BIM weapon of choice, as an online service, or as a bespoke customisation.
BIM 2.0’s killer feature is not going to be cloud; it’s going to be automation. Auto-drawings won’t just improve the speed of drawing production. They will ultimately mean fewer skilled people being tied up in documentation. If used well, this technology will also mean fewer licences of expensive CAD and BIM software needed in each firm. However, I am sure there will be a new business model from software developers eager to ensure that revenue isn’t lost in return for the productivity gains delivered. The good news for customers is that fierce competition means there will be plenty of options.