From Pretty Pictures to Real Progress: Climbing the AI Curve in AEC

Hello fellow magicians,

I’ve been attending many conferences and speaker series about AI in AEC lately, and I’ve noticed something.

So many speakers are brought in to talk about how they’re “using AI” in their architectural workflows. But the conversation tends to focus on just a few things:

• AI image generation in competition work or early design inspiration

• The general use of ChatGPT within the organization

• And inevitably, data quality issues

Here’s the pessimistic take: none of these are really examples of AI being integrated into architectural workflows.

Image generation is close, but let’s be honest—anyone who has tried to use it in a real project knows it’s mostly limited to “precedent” and “inspiration.” It doesn’t generate architecture. It doesn’t resolve constraints, coordinate systems, or solve problems. Similarly, ChatGPT is great for summarizing meeting notes or writing reports—but that’s operations, not architecture.

So what’s holding AEC back?
Why haven’t we seen the kind of deep, domain-specific breakthroughs that AI has enabled in industries like healthcare, finance, pharmaceuticals, or retail?

Let’s zoom out and look at how AI transformation usually plays out.

The AI Capability Curve (And Why It Matters)

Before AI starts delivering real value, industries tend to follow a familiar pattern. Let’s call it the AI Capability Curve—a five-stage path from awareness to impact:

1. Awareness
   → Realizing AI could reshape your work (design, estimating, scheduling, etc.)

2. Observation
   → Capturing how processes actually work—through logs, sensors, BIM, etc.

3. Aggregation
   → Building clean, consistent datasets across teams and tools

4. Modeling
   → Training machine learning models to solve high-impact, domain-specific problems

5. Impact
   → AI systems deliver measurable value in speed, quality, or insight

Each step builds on the one before it. Most of the real progress happens well before the modeling begins. And skipping straight to the shiny stuff usually ends in frustration.

How Other Industries Climbed the Curve

If it feels like AEC is lagging behind other industries, that’s because it is—but not in potential. It’s just earlier on the curve. To understand where we’re headed, it’s worth looking at how other data-heavy industries navigated the same maturity path—and what had to happen before AI really changed the game.

💳 Finance: Decades of Structured Data, Then Breakthroughs

Financial institutions were among the first to adopt AI—not because they were especially innovative, but because they had the prerequisites:

• Highly structured transactional data going back decades

• Regulatory pressure to log and monitor decision-making

• A culture of statistical modeling (think: actuarial science and quantitative finance)

Even with that foundation, it still took years of effort:
Banks built centralized data lakes, cleaned legacy systems, and hired PhDs and ML engineers to start automating fraud detection, credit scoring, algorithmic trading, and compliance monitoring.

The key? They had a history of clean, labeled, consistent data.
AI didn’t just land in a spreadsheet—it was trained on deep institutional memory.

AEC parallel: We don’t yet have a “transaction log” of how design evolves or how coordination issues are resolved. Until we start capturing that, AI can’t learn from it.

🧬 Healthcare: The Long Road to Diagnostic AI

AI breakthroughs in healthcare—like detecting cancer in radiology scans—made headlines. But what rarely gets discussed is what made them possible:

• The digitization of electronic health records (EHRs)

• Medical image standardization (DICOM format, etc.)

• Shared clinical datasets across hospitals and research institutions

• Cross-disciplinary collaboration between clinicians, data scientists, and software vendors

Even then, progress was slow. Clinicians were rightly skeptical. Models had to be interpretable, not just accurate. And strict validation protocols had to be developed before AI could be trusted in real-world settings.

The result? AI is now a diagnostic partner, not a replacement—flagging abnormalities, predicting disease progression, recommending treatment paths.

AEC parallel: Like doctors, architects and engineers work in high-risk, highly contextual environments. AI will need to earn trust slowly—by surfacing options, highlighting risks, and supporting decisions rather than replacing them outright.

💊 Pharma: Accelerating Drug Discovery with Generative Models

Pharmaceutical companies are now using AI to design entirely new molecules—compressing what used to take years into weeks or months - and winning the 2024 nobel prize in chemistry.

This wasn’t luck. It followed years of investment in:

• High-throughput screening platforms

• Digitized lab workflows

• Centralized compound libraries and chemical databases

• Collaboration between biologists, chemists, and machine learning researchers

With that foundation, companies trained generative models to explore chemical space more effectively than humans could—proposing drug candidates that are now moving into clinical trials faster and more efficiently.

The payoff?

• Faster time to discovery

• Lower failure rates in early-stage development

• AI-assisted repurposing of known compounds during crises (like COVID-19)

AEC parallel: Imagine generative models proposing detail assemblies or material systems based on firm preferences, climate data, and project type—automatically filtered for code compliance and constructability.

🛒 Retail: From Clicks to Prediction Machines

Retail has become one of the most AI-mature industries—not because of its design complexity, but because of its sheer data volume and feedback loops.

Every click, scroll, cart addition, and return generates structured behavioral data. That means retailers can:

• Predict demand

• Optimize pricing dynamically

• Personalize recommendations

• Forecast inventory and supply chain risk

But that transformation took years. It required:

• Data centralization across physical and digital storefronts

• Investments in cloud infrastructure

• The hiring of data science teams, product managers, and ML ops engineers

Today, AI powers everything from what items show up in your app… to how warehouses restock shelves overnight.

AEC parallel: Imagine if firms tracked how design elements were used across projects, which assemblies led to more RFIs, or how layout decisions impacted performance. With that level of connected data, AEC could begin training models that support the design and delivery lifecycle just as retail supports consumer behavior.

🚧 None of This Was Easy

In every case above, the payoff only came after deep investment in:

• Data readiness

• Workflow instrumentation

• Cross-disciplinary collaboration

• Patience

There were no shortcuts. But those who committed to building the pipeline saw exponential value once AI models were introduced.

AEC isn’t behind because it’s slow—it’s behind because its data has never been productized.
The challenge now is doing the unglamorous work of capturing, structuring, and contextualizing how we design and build.

🧱 We’re Still Laying the Footings

Looking across industries, one thing becomes clear:
AI doesn’t transform an industry overnight—it transforms the industries that are ready.

Finance, healthcare, retail, and pharma all followed the same arc. Not by luck, but by investing early in data infrastructure, standardized workflows, and the connective tissue between domain knowledge and technical execution.

AEC is on that path—but we’re not at the tipping point yet.

Right now, most of the industry sits somewhere between Stage 1 (Awareness) and Stage 3 (Aggregation) on the AI Capability Curve.

• We’re excited.

• We’re experimenting.

• Some firms are starting to structure and centralize data.
  But large-scale, model-driven impact—the kind that reshapes workflows or unlocks new value—is still rare.

And that’s okay. Because this is how it starts.

If we want to move from image generation and meeting summaries to real architectural intelligence, we have to keep doing the slow, foundational work:

• Capturing what happens during design

• Structuring and normalizing data across tools and teams

• Creating feedback loops that teach us what worked

• Building trust between designers, engineers, and the models they might someday rely on

It’s not glamorous. But it’s how every AI breakthrough has ever happened.

And here’s the strategic reality:
The firms that reach the top of the curve first won’t just be more efficient—they’ll operate on an entirely different level.

They’ll design faster. Coordinate smarter. Predict problems before they surface.
They’ll build institutional knowledge into tools that scale.
They’ll win work not just by design quality, but by intelligence, repeatability, and speed.

So here’s the question I’d pose to all of us in the AEC+AI space:
What are you doing now to move your firm—or your data—from Stage 2 to Stage 4?

That’s where the real leverage lives.
And that’s where the next wave of architectural intelligence will begin.