DC

It’s a good day when more steel is made in America

Spoke with the owner of a steel erection company Friday. Large company has been building in NYC on large scale infrastructure longer than I’ve been alive Spoke about steel pricing today and the speculative nature of it is making projects incredibly difficult. Something he has never seen in his life before Seems like the old playbooks are chopped Physical constraints are much harder to overcome

When intelligence diffuses throughout the construction industry we are going to be able to build the coolest shit

I gave Astra-6 site photos, engineering drawings, and access to Revit. 40 minutes later, it modelled it in Revit. It's changing how I do engineering. The gap between “I need this modeled” and “the model exists” has collapsed dramtically.

AI is stress-testing every bottleneck between ambition and deployment. Building standards won’t be exempt forever.

Would be nice if Bethlehem steel’s PA plant didn’t look like the rust map rn

Pablo on the most underrated run rn

When I tell my agents they aren’t citing the building code correctly

Giving Weimar

At the point of no return the least we can do is have some fun

Suffolk and MIT released Construction in the Age of AI this week. It’s worth a read but here is my take. The study looks at six areas where AI could meaningfully change construction: design automation, offsite manufacturing, permitting, scheduling, skilled labor/subcontracting, and supply chain/procurement. What I find most interesting isn't really any individual use case. It’s what happens when these systems begin working together. Design becomes less sequential. Right now we design something, analyze it, coordinate it, estimate it, revise it and then repeat that process over and over again. I think we're moving toward a world where we can continuously evaluate thousands of possible designs against engineering performance, cost, schedule, constructability, code requirements, procurement constraints and whatever else matters to the project. This inherently changes scheduling. A schedule stops being a mostly static representation of what we think will happen and starts becoming a dynamic system that continuously responds to design changes, labor availability, procurement, site conditions and actual progress. Eventually I think designing infrastructure starts looking a lot more like designing planes where we simulate far more of the system before we ever physically build it. But there is still a massive problem underneath all of this. Data. AEC data is shit frankly. Drawings are PDFs. Specifications are documents. Codes and standards live in different systems. Cost data lives somewhere else. Schedules have their own structures. BIM models have another. Manufacturers describe products differently. Contractors, engineers, owners and software platforms all organize information differently. Even when the information exists digitally, that doesn't mean machines can reliably use it together and AI doesn't magically fix that. You can build incredibly capable reasoning systems, but if the information entering those systems is incomplete, incompatible, poorly structured or inaccessible, the intelligence sitting on top of it eventually hits a ceiling. For decades we could tolerate this because humans were basically the interoperability layer. An engineer could open a PDF, interpret a table, understand a note, look something up in another document and manually translate that information into whatever format the next person or software needed. Agents don't eliminate that translation problem. They expose how big it actually is. If we want autonomous design, continuous scheduling, automated permitting, procurement optimization and eventually highly simulated construction, machines need to be able to exchange engineering knowledge reliably. I don't think that necessarily means forcing the entire industry into one universal data standard either. The more practical possibility to me is infrastructure that can translate between different standards and representations as needed. Someone producing information may structure it one way. The engineer, contractor, software platform or agent consuming it may need it another way. Something needs to sit between the two and make that translation possible without losing the meaning or authority of the original information. Really interesting work from Suffolk, the MIT Center for Real Estate, MIT Media Lab City Science, and everyone from the industry who contributed to the research.

Over the past week, I’ve been rebuilding the original World Trade Center. The goal was never to recreate September 11. I wanted to remember the World Trade Center as it stood before that day. My own small way of making sure it is never forgotten. Remembering everyone we lost on September 11, 2001.

People are saying

I can say first hand Astra is really good in revit and unreal engine

Thank you @XMoney

I actually agree with the direction here unfortunately Nikita probably didn’t realize how much of a hornets nest the AEC industry is when it comes to these things Revit mcp has been very helpful for me drafting details and putting together plans Fighting the Astra rate limits but the gpt6 model seems very good at modeling/drafting

Request for more domestic steel

Love that @bot

This looks incredible

we should be able to do better than this