Key themes
Generative design
Great idea, long history of poor implementation. The trap is generating too many bad options instead of a few good ones.
Co-creation
The human reaches into the generated model and manipulates it. The computer resolves. Back and forth until the building works.
Deal opportunity cost cycles
Almost everyone in the early development process works on opportunity cost. The cycle repeats until someone gets happy — or the deal dies.
Design for documentation
Kit of parts in TestFit 2.0 links feasibility geometry to real Revit unit plans, closing the gap between massing and documentation.
The deal opportunity cost cycle
Most people in AEC don’t understand what a developer actually does. Here’s the full loop as explained in the episode:
1
Landowner hires a broker to sell a site
↓
2
Broker sends the site to a developer
↓
3
Developer hires an architect — often for free — to study the site
↓
4
Architect returns a scheme. Contractor prices it. Subcontractors price deeper.
↓
5
Developer takes it to investors and the bank. Bank says: “Can you do a little better?”
↓
6
Consultants get squeezed. More units. Better yield. Run the cycle again.
↓
7
Landowner reconsiders their ask. Everyone is upset. Run the cycle again.
↓
8
SDs. Engineers. Column in the wrong place. More iteration.
↓
9
Permit filed. City says no.
The further upstream you go, the less people actually know about buildings — even though buildings are the product the whole industry runs on.
— Clifton Harness
On generative design
We got feedback as early as 2016 that people would rather have three good options than thousands of bad ones. Humans don’t do well with option overload. We brood instead of work.
— Clifton Harness
The failure mode of most generative design tools is death by options, bad UX, and forgetting the human altogether. Clifton’s take: the human mind is still better at understanding complex problems simply and intuitively. Computers are better at math. That’s basically it.
The other trap is internal scripts. A firm spends three months building a Grasshopper workflow in a bad IDE, then has to maintain it across five changing platforms. Leadership then asks why it can’t also do twelve more things. The script “costs nothing” — except the thousands of dollars in time that nobody counted.
On architecture and automation
I’ve said this directly to the AIA: the profession will die if it doesn’t change. Architects need to own the process by becoming masters of the data — not just whether they can draw beautifully on trace paper.
— Clifton Harness
No one is trying to take away the joy of designing form. TestFit just removes the drudgery: counting units, calculating square footage, making sure rules are followed, and surfacing the constraints while you design.
— Nat McDonald
My definition of architecture is the meaningful creation of space. If it’s not meaningfully created, it’s probably not going to be good architecture.
— Clifton Harness
Revit integration workflow
Nat walked through the active integration work at the time of the episode. The goal was to close the gap between TestFit feasibility and Revit documentation without forcing users to redraw everything.
Property line → TestFit
Pull a site boundary directly from Revit into TestFit to start a scheme without retracing.
Revit units → kit of parts
Take real unit plans from Revit and turn them into a TestFit kit of parts, so actual unit logic drives the feasibility model.
TestFit → Revit geometry
Feed an RSD file through a Dynamo script, map building objects to Revit family types, and generate actual walls, floors, and roofs.
Units back into Revit
Place typical units back into the generated Revit building using model groups, positioned using demising walls from TestFit.
Building types covered
At the time of this episode, TestFit supported: high-density multifamily, surface and structured parking, office (with life safety and core logic), lower-density housing including townhomes and duplexes, and hotels with modular room planning. Structural column placement via Thornton Tomasetti’s API was newly live.
Full transcript
Guest introductions
Clifton Harness
I’m Clifton Harness. I live in Dallas. I convinced my college roommate, now our CTO Ryan Grieg, to help me automate parking garages. That’s really where I got started, along with my absolute hatred for counting parking spaces. That kicked off a four-year journey of asking: how do we disrupt the planning process, get people off trace paper, and move them into data-driven design tools?
Nat McDonald
I’m a licensed mechanical engineer. Over the past seven years I followed a pretty traditional engineering path at a variety of firms, with Buro Happold being the last one. During that time I picked up a lot of computational design knowledge and developed a real passion for it. I decided to make the jump into software and join TestFit. It felt like a place where I could really scale that design knowledge.
Jeff (The Revit Kid)
So you went from ducts and diffusers to parking spaces?
Nat McDonald
Yes. And I’m so over ducts. They serve one purpose: bring air from one spot to another.
Generative design: a good idea with bad implementation
Clifton Harness
Generative design is a great idea with a long history of poor implementation. A lot of current tools have terrible UI and UX. You draw a rectangle by wiring up points and nodes, extrude it, get a box. The box has no real information in it. That’s not useful enough for most people in AEC.
Clifton Harness
So what happens? Firms build giant scripts in a bad IDE, spend three months on it, then have to maintain it while five different applications keep changing. Leadership says, “We spent three months on this. Why can’t it also do twelve more things?” That’s the trap.
Clifton Harness
One of the biggest failures of a lot of generative design tools is death by options. We got feedback as early as 2016 that people would rather have three good options than thousands of bad ones. Humans don’t do well with option overload. We brood instead of work. The other damning thing is that a lot of current generative design forgot the human altogether. The human mind is still better at understanding complex problems simply and intuitively. Computers are better at math. That’s basically what they’re better at.
Clifton Harness
So our approach was different. We wanted something intuitive, supported, paid, broadly usable, and focused on commodity building types: apartments, office buildings, hotels, parking garages. Quality options over quantity. The core feature we landed on was co-creation.
The real problem: deal opportunity cost cycles
Clifton Harness
The process starts with a landowner. The landowner hires a broker. The broker sends a site to a developer. The developer wants to buy the land and put a building on it for profit. To do that, they need consultants: architects, contractors, subcontractors, manufacturers, bankers, investors. That whole ecosystem gets activated around a site.
Clifton Harness
The developer looks at the site and says, “I don’t know how to draw buildings,” so they hire an architect. In the U.S., the architect is often doing this early work for free in hopes of winning the rest of the job. The architect sends something back. The developer gets pricing. All of this goes back up. But the developer is usually only about ten percent of the money required to build the building. The rest comes from limited partners, investors, and debt.
Clifton Harness
So the bank says, “This looks good, but we think you could do a little better.” The developer squeezes the consultants again. More units. Better economics. Run the cycle again. Then maybe everyone gets happy and it goes back to the landowner. The landowner says, “Actually, I thought I was getting two million, not 1.8.” Run the cycle again. Then the project goes into SDs. A structural engineer puts a column in a place the architect hates. More iteration. Eventually you file a permit set. Then the city says no.
Clifton Harness
That’s the ecosystem. Almost everyone is working on opportunity cost. Very few people are actually getting paid well during these early loops. Our view is that TestFit is well positioned to disrupt this process — and the further upstream you go, the less people actually know about buildings, even though buildings are the product the whole industry runs on.
Architecture, automation, and design control
Jeff (The Revit Kid)
A lot of architects hear this and worry that the software is taking away the part they care about: laying out the building. What’s your response to that?
Clifton Harness
We started with parking garages. I don’t know a single architect who is truly passionate about laying out parking stalls over and over. A lot of the same people worried about automation are also the ones saying someone should do something about the housing crisis. There’s a lot of hypocrisy in AEC when it comes to technology.
Clifton Harness
Architects traditionally have been the gatekeepers of spatial thinking, and they’re good at it. But if they don’t change how they monetize their value and make themselves essential to the process in a new way, architecture as a profession is going to keep losing ground. I’ve said this directly to the AIA: the profession will die if it doesn’t change.
Nat McDonald
No one is trying to take away the joy of designing form. TestFit just helps by removing the drudgery: counting units, calculating square footage, making sure rules are followed, and surfacing the constraints while you design.
Jeff (The Revit Kid)
That’s why I like the term co-creation. Even if I enjoy sketching and massing buildings, I’m still designing within a set of rules. The software just makes those rules visible and responsive.
Clifton Harness
Exactly. My definition of architecture is the meaningful creation of space. If it’s not meaningfully created, it’s probably not going to be good architecture. That was part of my issue with early generative design — it often wasn’t meaningful. It generated a bunch of junk without embedding actual design intelligence.
Kit of parts and design for documentation
Clifton Harness
With TestFit 2.0, we introduced the concept of KOP, or kit of parts. In the early days, we generated units as simple boxes with areas attached. In 2.0, we started generating with more fleshed-out parts — units can now have associated unit plans. Ideally, those unit plans already exist in Revit from a previous project or from the project team. We don’t want to be in the business of authoring final unit plans ourselves. That’s still core architectural territory.
Clifton Harness
The software distinguishes between inline units and corner conditions. Inline units are the majority. Corner units are trickier. When the building mass gets weird, they need to respond to odd geometry — that’s where the algorithm tells the user: “You need to design this part.” That’s still co-creation. The computer handles the boring solving, and the user meaningfully creates the building.
Real-time deal prototyping
Clifton Harness
One of the most interesting workflows we’ve been training architecture customers on is real-time deal prototyping. Imagine a Zoom call where all the stakeholders are present: the architect, the developer, the contractor, maybe even the owner. The developer can plug in assumptions — below-grade parking at $45,000 per stall, above-grade at $22,000 — and see the building respond in real time. You can hear the developer start to articulate what they actually want out of the deal. They start getting excited about building design in a way they often never have before.
Revit integrations
Nat McDonald
We’ve been working on quite a bit here. I built a tool to take a property line from Revit into TestFit. I’ve also been building workflows that let you take Revit unit plans and turn them into a TestFit kit of parts. And we’ve built ways to translate TestFit geometry back into BIM geometry.
Nat McDonald
Right now the prototype workflow uses Dynamo, though the long-term goal is to wrap this into an add-in so users don’t have to deal with Dynamo spaghetti. The user feeds in an RSD file — the TestFit save file — maps building objects to Revit family types, and the script runs and generates actual Revit walls, floors, roofs, and openings from the TestFit scheme.
Jeff (The Revit Kid)
That’s a big deal. Normally someone would still be redrawing the building in Revit. This gets you much further, much faster.
Nat McDonald
Exactly. And another workflow lets you take typical units from Revit and bring them into TestFit, then place them back into the Revit building using model groups positioned from TestFit’s demising walls. That does raise questions about model heaviness — large Revit files with lots of repeated groups can get slow fast — so we’re also thinking about better ways to reference typical units to avoid that overhead.
Structural integration
Clifton Harness
One of my favorite recent features is our structural integration. We heard early on that users would say, “My stall average is off. Where are the columns? This isn’t a realistic garage.” So we worked with Thornton Tomasetti and their structural AI tools. Now we can activate a column buffer, send the building logic to their API, and get column sizing and placement back in the model. We’re not just automating geometry — we’re interacting with design knowledge from one of the best structural engineering firms in the country.