Following the last interview we did with Aaron Woolverton of Design Workshop about their use of digital tools, this week we’re sharing a discussion with Kristen Padavic of Cedar. Headquartered in Austin, TX, Cedar looks kind of like an architecture firm and kind of like an AECO technology company. It’s both, and while that may feel a little unusual, it is a keen response to the state of technological development right now. While the latest generation of AI is already great at coding, that’s because there’s a wealth of technical training information on the open internet: tutorials, forums, videos, you name it. For architecture and urban planning, the open internet is just not as jam-packed with useful data on which to train LLMs. Thus a window of opportunity opens, and Cedar is enjoying the view. More after the jump.
💬 Hello! This is the newsletter of the Urban Technology program at University of Michigan, in which we explore the ways that data, connectivity, computation, and automation can be harnessed to nurture and improve urban life. If you’re new here, try this short video of students describing urban technology in their own words or this short explainer video.
👷♀️ Eat Your Own Dog Food Before AI Eats Your Lunch
When a talk starts with “I like to call myself a full stack architect,” you know that there’s something interesting to follow because the speaker, in this case Kristen Padavic of Cedar, is able to natively bridge the urbanist/technologist divide. Kristen is an alum of Taubman and she stopped by the UT studio on a recent visit to campus.
Cedar describes themselves as an “architectural design firm built to partner with the modern real estate developer” that has created a platform for evaluating sites and designing housing. In my own words, they combine the feasibility function of the Testfit / Digital Blue Foam / Delve (now part of Google Earth) type products with a set of architectural designs for ‘housing as a product,’ kinda sorta like Juno, and in-house design expertise.
Have you ever tried to pour peanut butter out of a jar? That’s about how fast the real estate-to-architecture-to-construction pipeline often feels. Cedar uses AI/ML to speed up site selection and feasibility analysis in a holistic way that combines zoning, tree coverage, topography, and so forth: the kinds of things you need to know about to really understand exactly what can be built, sure, but more importantly what can be built at reasonable cost. Kristen describes the benefits here as accruing to efficiency and quality both: “80% of what architects do is rote, rule-based crap. If we can automate that, then we get to spend 100% of our time being great designers.” The lesson here is to automate typical situations so you can embrace the edge cases. That lines up with our perspective that the AI era makes judgement even more important. Judgement is the last mile of decision-making.
The result is that they are architects who work at the speed of developers, allowing them to partner with developer clients in ways that previously would have been serviced by a small cast of discrete consultants. Cedar’s internal tooling lets them work so fast they can just go ahead and calculate in advance how many sites there are in Austin where their housing products are viable (18,000) and same for all of Texas (1M). Because they’ve designed the housing products as a suite of different options, they know that on these 1M sites it’s not just theoretically possible to build multifamily housing, but that they could file for a permit to do so tomorrow.
As Kristen explained, this kind of business was unlocked by a confluence of technological and urban trends. The tech side is easy: first ML and then AI have made possible the complex analysis this kind of work demands. The urban side is about zoning changes that open up new places to build housing. This includes changes such as up-zoning, ADU laws, and removing parking minimums. In sum, these changes mean infill development is an area of expanding opportunity in many communities around the country, but/and that infill development is harder to execute. The feasibility is trickier and the design more challenging than doing the equivalent project on virgin land somewhere on the sprawling fringes of town. Cedar’s bet is that the extra complexity that comes with infill development can be managed through well-designed software.
Usually developers live and breathe by the zoning ordinance and know very well how to squeeze the most out of it, but because of the recent wave of zoning changes, Cedar finds that many developers are still catching up. Cedar’s tooling makes it irrelevant: just use the model to test out scenarios on a site. In fact, they can test building scenarios the client wasn’t even asking for, such as single stair buildings (multi-unit apartment buildings with one fire stair instead of the 2+ typically required) which are newly enabled by code changes in some jurisdictions. Because of that newness, single-stair buildings can be perceived as risky, so Cedar’s proactive exploration of analyzing sites for their ability to embrace an important code change in a way of de-risking new building types for developers. This is not a solution to affordable housing in the US, but it’s an ingredient for sure.
Software replacing consulting hours is important because it lets you explore scenarios rapidly. Who else can benefit from scenarios? Just ask my colleague Rob Goodspeed, author of Scenario Planning for Cities and Regions. Cities, of course, and Kristen is already doing so: “We [have worked] with city planning offices in Texas to help them contemplate code changes” to answer questions like “how much more housing would be allowable if we made this zoning change?”
At the moment Cedar does some feasibility analysis by hand using their internal design team, so they’re effectively building their own AI training set on actual projects. This reminds me of working in a Kleiner Perkins incubator space on Sand Hill Road back in the year 2000. It was there that I learned the phrase “eat your own dog food” as we tested the software we were building by using it first among ourselves in-house. Kristen’s work shows that a new aphorism is necessary: it’s time to train your own dog. There’s a race on for who will build the most capable models. As Anthropic, OpenAI, Meta, and Google duke it out for general purpose models trained on the open internet, domains with deep requirements like architecture and planning are not being served very deeply—though it’s worth noting that this could change soon, as Autodesk is working on a foundational model for AEC.
There are not many companies that embrace technological tool building and design work in the way that Cedar does. This blend of eating your own dog food and training your own dog makes so much sense, but is exceedingly hard for a pure consulting firm (which is what most architecture firms are) to find money and time to dedicate to innovation work. “Everything about architecture is about the billable hour,” Padavic explains, “which is why it’s tough to innovate. If it’s not directly relevant to the project, you can’t bill [the client] for that! You can innovate within a project, but not general purpose exploration” and product development.
Ironically, the business model of architecture and planning firms is a barrier to innovating in how architects and planners work. To create radically better ways of working it requires new talent (engineering/dev experience), new business model (software vs. consulting), and a legitimate collaboration of urbanists and technologists, and Cedar is doing exactly that.
Thanks for visiting, Kristen!
These weeks: UT++ drafts. Light work on project BICOE. Got the vestaboard working again! 🏃