Agents, not Features: Rethinking AI in Design Tools
Key Takeaways
Agents vs. Features: AI should be viewed as a collaborative teammate (agent) that reasons and acts, rather than a static tool or template.
Multiplayer Integration: AI agents can join files like human collaborators to observe and assist in the design process.
Context-Aware Automation: Future agents will leverage a firm's historical data to build presentations and reports based on previous design languages and constraints.
Practical Application: Arcol’s first agents plan to focus on "boring but powerful" tasks like automated costing by pulling data from internal databases and web sources.
The Goal of AI: True AI in architecture should focus on collaboration and embedded intelligence rather than just being a "black box" generative tool.
How would this work?
Imagine you’re working on a new building concept. You’ve mocked up your model, pulled together some metrics, and now it’s time to build a presentation. Normally, you’d start from scratch. You’d be dragging in screenshots, updating charts, formatting text boxes, pulling CSV files.
But what if you didn’t have to?
“What if I had a presentation agent that built it for me? Based on how we’ve built decks at your firm over the last decade?”
That’s what an agent does. It’s not just a feature. It’s a collaborator.
Because Arcol is multiplayer and real-time, an agent can join your file just like a teammate. You see it work. It sees your model. It reasons, interprets, and acts, just like you would.
“Why does it have to be a person behind the cursor? Why can’t it be a script I’ve written that goes and does reasoning for me?”
We think that’s the future. Not fiddling with more sliders. Not choosing from more templates. But asking, “Build me something like we did last time but make it work for this site.”
We're already building our first generation of these agents, starting with something boring but powerful: costing.
The costing agent automates the reporting process by analyzing:
Key Inputs: Location, typology, and square footage.
Data Sources: Internal cost databases and external Google search results.
Output: A fully formatted, automated cost report.
It's simple. It’s imperfect. But it's useful. And more importantly, you didn’t have to do it yourself.
We evaluate AI utility based on two primary criteria:
Efficiency: Does it save the user significant time?
Value: Does it think in a way that feels valuable rather than "uncanny"?
The Road Ahead
Will we build more generative tools? Sure. But we won’t call them AI if they’re not.
What excites us more is the idea of embedded intelligence that works with you, seamlessly. Agents that build, test, adapt, and learn from how your studio actually works.
We want to support a future where AI isn’t a black box. It’s more of a visible teammate with a lot of context. One that understands your design language, your constraints, your aesthetic, and your goals.
Because real AI in architecture shouldn’t just generate. It should collaborate.