Margaret Gereghty

I chose to watch the talk titled “The Future of BIM Is NOT BIM, And It's Coming Faster Than You Think - The Sequel” by Bill Allen. I found the speaker very engaging and overall enjoyed his positive outlook on the future of AI/ML in the AEC industry. In a time when many people worry that AI will make their jobs obsolete, Allen presented a refreshingly positive and realistic perspective.

Allen emphasized that AI/ML can function as a connection point between generative modeling and generative design. He stressed the importance of human involvement in the generative design process, referring to it as “co-authoring” with the computer. In this process, the human designer still retains the ability to evaluate and choose between design options rather than simply accepting whatever the software produces.

I also appreciated how he acknowledged that technologies can fail. He shared an example of a technology he attempted to implement at his own company that ultimately created more work instead of reducing it. Stories like this are important because technology should make designers’ lives easier, not harder, and developers can sometimes lose sight of that goal.

Because the talk was given in 2019, some of Allen’s points feel somewhat dated today. He mentions that AI/ML was not yet ready to handle some potential uses in the AEC industry, but current technology has advanced significantly since then. Additionally, some of the companies he referenced have since gone bankrupt, partly due to the COVID-19 pandemic but also because of how difficult it can be to introduce lasting technological change within the AEC industry.

In previous internships, I noticed that interns are often assigned simple, repetitive tasks that could likely be automated using tools such as Dynamo, including sheet creation, importing details, and modeling existing structures. After watching this talk, I think it would be very interesting to explore ways to automate some of these processes, similar to the ideas Allen discussed.

One point at which I found myself slightly limited by the tools at hand was in the last assignment, module 7, where we were tasked with developing a generative study to evaluate a given “problem”. I chose to analyze the interaction between structural bay size and available architectural room, but I found myself thinking that I wasn’t getting a realistic picture of the problem space and would have been better off with just knowledge from experience.

One way I think AI could be used to mitigate this fact is by building a code knowledge base. All engineers know that the codes are long, complex and incredibly daunting for young engineers. Specifically in ASCE 7 there is a lot of caviats based on structural material, loading, location, etc. I think it would be incredibly useful for an AI bot to generate/find those requirements outlined in the code and relate them to a structural bay study to get a better/more realistic idea for the true behavior of different materials at a similar bay size. I think pushing this further you could feed AI spec manuals from companies and have it store that data for engineers to reference in their generative studies as well.

Tool 1: Chaos

Chaos is a visualization company that mainly creates and sells architectural visualization tools. They have emphasized the use of AI/ML at “every stage” of the design process. Their architectural tools came from the acquisition of EvolveLab the company Ben Allen worked for at the time of his presentation at Autodesk University (2019).

They have integrated AI into their visualization workflow by allowing users to enter text prompts to change the visualization while still maintaining the original design intent. This means that users can generate many more renderings in 2D and 3D in a much shorter time frame and earlier on in the project phases. This could be incredibly useful for architects, who typically spend hours on renderings, and owners who can get a better feel for their project earlier into design, hopefully reducing rework and client unhappiness at later stages.

This company functions entirely separate to the visual programming tools I used this quarter so it would not have changed how I did my work, but would be very interesting to use. I have had experience with other AI generation renderings and I have been disappointed by them in the past, I think the direct integration with Revit and other softwares puts this company at the forefront for visualization in the AEC industry.

Tool 2: NomicAI

Nomic advertises itself as a customizable AI that is built to have the capability to read and analyze long and technically complex engineering documents and plans. The customization allows users to create agents for repetitive tasks like RFI and RHP response and code compliance, things that usually take engineers “up to 20 hrs per week”.

The agentic part of this AI bot it was separates itself from standard ChatGPT or Claude. It can be customized by the user to create additional interfaces for analysis and task delegation. Furthermore, witch access to all files at the company it can be a source for question and answer for new engineers as well.

This would not have changed the way I worked this quarter as well but it is an interesting concept. Reading about it left me wondering about how effective this tool would really be. All outputs would have to be verified by a senior engineer or architect like the standard engineering process. I think in a real world design setting it could see the most impact with younger engineers, because having a chatbot to ask all the silly questions that you have early in your career would be incredibly helpful (albeit reduce your interaction with other people, which is not great!).

Tool 3: BEAMAI

BeamAI is a takeoff and estimating software that leverages AI to create schedules and cost estimations in excel by reading plans in PDF form. The process is simple, you upload the plans and define the scope of work and submit. It outputs AI generated takeoffs and cost estimates (that are checked by an expert before being distributed) to the customer.

This could help construction workers astronomically, doing quantity takeoffs and reading plans can be a huge time sink on a project, especially when a CM does not have access to a BIM model (as it is in the real world most often). This would cut construction overhead costs considerably and allow CMs to focus on other aspects of the job.

This again, unfortunately, would not have changed the way I worked this quarter but was very interesting to read about. As an intern I spent a lot of time reading and modeling existing building plans so it made me wonder how this concept of reading plans could be applied to that task as well.