Part 1

Allen's main point is that the real problem is friction, not modeling. He uses the music example, going from driving to a store to buy a CD all the way to asking a voice assistant to play a song, and argues the same kind of collapse is coming for how we design and build. He sees BIM as a base layer that feeds optimization, fabrication, and eventually robotics on site.

What surprised me was how honest he is about computational design not being the answer on its own. He basically says generating thousands of options just creates a new problem because now someone has to pick one, and nobody actually evaluates all of them. I think that is the most useful thing in the talk because it admits optimization without judgment just moves the bottleneck.

The thing I would want to try is the test fitting workflow, since it lines up with the feasibility work I care about on the real estate side. For me to actually use it on a project, the output would have to be trustworthy enough to put in front of a client or a lender without me redoing the whole thing by hand. Otherwise the friction has just moved somewhere else.

Part 2

The hardest thing I did all quarter was rotating panels based on their relationship to the sun. Getting them to actually respond to a sun vector instead of just sitting at a fixed angle took forever, and every small change meant re-checking whether the rotation was reading the sun position right or just looking like it was. The most repetitive thing was building the H shaped building form, which was mostly redoing the same move over and over to get the wings and the connector to read as one mass.

For the panel problem I would want a tool where I describe the goal, something like keep these panels oriented to maximize sun exposure across the day, and have it set up the rotation logic so I can spend time judging the result instead of wiring up the mechanics. I would actually want that. For the H building I am less sure. Struggling through it by hand is part of how I learned what the geometry was doing, and I think handing it off too early would have left me with a model I did not really understand.

Part 3

TestFit is a real estate feasibility tool that generates building and site layouts from inputs like the parcel, setbacks, unit mix, and parking, with the yield already calculated. I think it is interesting because it attacks the gap between a napkin sketch and a real feasibility study, which is usually slow and manual. It would have helped me more on the real estate side than the parametric side, because that early feasibility loop is where I lose the most time.

Maket.ai generates floor plan layouts from parameters and constraints like site size, setbacks, and design rules. What makes it new is that it treats layout generation as something you prompt rather than draft. It would have helped with the repetitive H building modeling because the part that felt like a waste was generating and regenerating the basic form.

The third is an AI tools plugin for Grasshopper that adds a text to CAD component, which generates geometry inside Grasshopper from a text prompt, and an image to prompt component that turns a reference image into a prompt you can feed back in. I think this one is the most relevant to the class because it lives inside the same environment we used all quarter. It would not have solved the sun rotation problem since that is logic, but it might have given me a faster starting point for the H building