Part 1 - Study one talk
Picked Option B - The industry vision: The Future of BIM Is NOT BIM, And It's Coming Faster Than You Think - The Sequel (Bill Allen, AU 2019)
Bill Allen's talk starts with a problem: the AEC industry generates enormous amounts of valuable data and design potential, but mostly leaves it on the table. His argument is that we are moving out of Building Information Modeling and into Building Information Optimization and Automation, with what he calls a "messy middle" in between where teams still brute-force results with Dynamo and Grasshopper scripts. The approach he proposes is to stop designing "passively," one tired human plus one computer producing limited options, and instead feed the computer rules, minimums, and maximums so it can return thousands of optimized solutions. He walks through optioneering with Project Refinery (using a Mod Pizza analogy), evolutionary problem solving with Galapagos for a Dubai master plan, and generative tools like TestFit and a Revit test-fitting tool. What surprised me most was how far computational optimization has already moved past the slide-deck stage and into real practice. I expected generative design to still be at the forefront, and Allen himself admits it often is, but the examples were concrete: Mortenson's solar field optimizer, Brett Young's AI solver that ran 1.2 million coordination iterations to lay out MEP runs, and a modular panel project where every unique panel carried its own QR code for assembly, "like IKEA for buildings." Seeing optimization tied directly to industrialized construction, packing algorithms for shipping, auto-generated shop drawings, and robotic bricklaying with SAM made it feel less like a prediction and more like a status report. The piece I would most want to try is the dynamic data dashboard work. The idea of pulling clash detection, trending, punch lists, and project profitability into one live, physics-driven view, rather than static schedules that nobody reads, seems immediately useful. For me to actually build it on a project, a few things would need to be true: the source data would have to be reasonably clean and consistently structured across the model, there would need to be buy-in from project managers who would actually use the metrics, and someone on the team would need enough scripting comfort to maintain it. Beyond dashboards, I would want to test computational optimization on a repeatable construction problem, something like layout or sequencing, where efficiency gains are measurable.
Part 2 - Reflect on the quarter
Looking back at the parametric work in Dynamo this quarter, two friction points stand out. The first was node recall. Every time I built a graph, I found myself stopping to remember which node did what, or digging back through an old example file to copy the right setup. That constant context switching broke my flow and took more time than the actual design thinking did. The second was the test-run loop. I'd wire up a chunk of logic, hit run, watch it fail or return nulls, adjust one node, run again, and repeat. Flexing a form to evaluate metrics often meant over twenty runs before it behaved, and most of that was guesswork. An AI-augmented version could close both gaps. I imagine describing what I want in plain language, "enable to flex this tower's floor plate and plot daylight against floor area," and the tool proposing a node graph as a starting plan, the way vibe coding works. Better still, it could run and debug itself, catching the null upstream and suggesting the fix before I ever hit run. I would actually want this augmentation. The friction here wasn't a meaningful struggle; it was busywork. If a tool is too complicated to remember, that complexity isn't allowing me to retain other than the concepts, and would not be fully utilized in the industry.
Part 3 - Scout the frontier
Aurivus (AI scan-to-BIM)
Aurivus is a tool that automates the conversion of laser scans into usable BIM. Aurivus automates the scan-to-BIM process by first reading 3D laser scans as point clouds and then analyzing them using AI. The AI recognizes components such as walls, pipes, or beams and provides them as structured objects. The problem it tackles is that turning a raw point cloud into a modeled as-built has historically been slow, manual, and modeler-dependent. Using computer vision to classify elements and snap them to families is what wasn't really usable a decade ago. My coursework didn't involve existing-conditions capture, but if a future project starts from a real site rather than a clean massing, this is the bridge between reality and a parametric model.
Buildots (AI progress tracking from helmet cameras)
Buildots is a construction tech platform that uses computer vision to track what's actually being built and compare it against the BIM model and schedule. Rather than relying on dedicated scanning crews, their system captures images from hard hats worn by workers or mounted cameras around the site, processing these visuals with sophisticated algorithms to provide real-time updates on project status. The data flows two ways: it has a two-way integration between our platform and the leading scheduling software applications in the industry, which enables us to both retrieve the most up-to-date schedule, and more importantly, to feed back the latest progress, syncing automatically with tools like Primavera P6, Asta, and Microsoft Project. Founded in 2018, the company recently raised a Series D, expanded into underground utility tracking from drone footage, and added a generative AI assistant called Dot for querying site data.
Trunk Tools (AI agents for construction documents)
Trunk Tools is an AI platform that ingests the mountain of paperwork generated by a construction project and makes it answerable in plain language. It uses advanced AI to organize and provide insight into the millions of documents, drawings, specs, RFIs, schedules, and submittals generated throughout the lifecycle of a project, letting field teams ask a question and get a sourced answer instead of digging through PDFs. Its chat agent, TrunkText, goes further than retrieval: powered by a large-language model, it can even automate workflows like comparing a submittal against project specifications, the kind of document cross-checking that used to tie up a person for hours. Founded in 2021 by Dr. Sarah Buchner, it raised a $40M Series B in July 2025, bringing total funding to $70M. What makes it interesting is the platform’s framing as an active teammate rather than a passive file store. The pitch is essentially "ChatGPT for construction sites," but the founder’s distinction is that general-purpose AI struggles here because the industry runs on proprietary data, dense jargon, and tangled workflows, so the model has to be grounded in structured project data to be trustworthy. A worker can ask something as specific as whether a given door requires electricity and get an answer with the source drawing attached. The scale of the problem is real: one example project involved roughly 21,000 documents and over 33 GB of data.