CA in the Age of AI: Faster Isn’t the Point

AI promises in architecture are getting bigger by the day.

“10× faster submittal reviews.”
“RFI responses in seconds.”
“Instant spec compliance checks.”
“Cut CA time by 80%.”

Speed has become the default way we talk about AI. And it makes sense. Construction administration is full of work that takes too long, and nobody is asking for slower submittal reviews.

But faster isn’t really the point.

Because when you look at where time is actually being lost during CA, a different picture emerges.

The opportunity for AI isn’t simply to help architects type faster. It’s to remove the administrative work surrounding decisions, so architects have more time to make the decisions that actually require their expertise.

The industry doesn’t have a writing problem. It has a waiting problem.

We looked at aggregate data from roughly 270,000 document reviews across North American CA projects between 2023 and 2026.

The median review took five days.

The average took 25.

That gap matters. It tells us there’s a very long tail of reviews sitting unresolved for much longer than they should.

And it isn’t one discipline causing the problem.

Across architectural, structural, mechanical, electrical and civil reviews, roughly half were late. Architectural reviews were late 48% of the time. Structural: 51%. Mechanical and civil: 52%. Electrical: 53%.

When everyone misses at roughly the same rate, it becomes hard to argue that one discipline is the bottleneck.

The system is the bottleneck.

Think about what happens between receiving a submittal and actually making a decision.

Someone receives it. Logs it. Figures out who needs to see it. Tracks down the relevant specification section. Finds the applicable drawings. Compares the submission against the contract documents. Coordinates comments. Follows up with consultants. Drafts a response. Updates the record.

Only some of that work requires architectural judgment.

And that distinction is where AI starts to become genuinely useful.

Let AI handle the busywork. Keep the accountability.

There are already several areas of CA where AI can do useful work today.

It can help triage incoming documents by logging, routing and tracking them.

It can handle retrieval, finding relevant information in the project record and, importantly, showing the source.

It can assist with comparison, checking a submission against the contract documents and flagging potential deviations for review.

And it can draft responses, giving the architect a first pass rather than a blank page.

What it can’t do is take responsibility for the decision.

Consider a submittal review.

AI might log and route the submittal in seconds. It can retrieve the relevant specification sections, with citations, in seconds. It can flag deviations and prepare a draft response in minutes.

Then comes the part that shouldn’t change:

The architect decides. The architect stamps.

That’s the line we need to preserve as AI becomes more capable.

AI can write every word. It can’t own any of them.

The goal shouldn’t be removing architects from CA

There’s a tempting version of the AI story where automation steadily removes humans from the process.

That’s not the future of CA we think firms should be working toward.

The better question is: What could architects do if the administrative burden of CA became dramatically smaller?

Maybe you could actually spend more time looking at the site — not just glancing at the latest field photos, but examining them closely enough to notice what’s happening in the background.

Maybe same-day submittal turnaround could become part of your service promise rather than an occasional victory.

Maybe instead of answering the fifteenth RFI caused by the same drawing issue, you could mine the project record, identify the pattern and fix it in the next set.

And maybe junior architects could spend less of their early careers formatting transmittals and maintaining logs, and more time sitting beside experienced architects while they make judgment calls.

That last one matters.

If AI takes away some of the repetitive work traditionally given to junior staff, firms will need to be much more intentional about how the next generation learns. The goal shouldn’t be to automate away apprenticeship. It should be to create more opportunities for juniors to learn the parts of the job that matter most.

Five questions to ask before AI touches your project

There’s another side to all of this: adopting AI simply because a vendor has added it to a product is not a strategy.

Architecture firms are responsible for their project records, their decisions and ultimately their work. Any AI system entering that environment deserves scrutiny.

Before adopting one, ask:

1. Show me what happens when it’s wrong. A useful AI product should be designed around the reality that AI will make mistakes, not pretend otherwise.

2. Do you measure time saved — or errors caught? Efficiency matters, but catching something a human might otherwise have missed can be far more valuable.

3. Where does my project data go? Firms should understand how their information is stored, processed and used.

4. What does this add to my workflow? An AI feature that creates another place to copy, paste, upload or verify information may simply be moving the administrative burden around.

5. What happens to my record if your company is acquired? Your CA record may need to outlive the software company storing it.

These aren’t anti-AI questions.

They’re the questions we should be asking precisely because AI is becoming important enough to take seriously.

Where CA goes from here

The most interesting changes may ultimately be bigger than faster document reviews.

The best RFI may be the one that never gets written.

As systems get better at understanding the project record, some routine questions could be identified and resolved before they ever become formal RFIs.

CA may increasingly become exception management.

Routine information can move from machine to machine, while architects focus their attention on the situations that actually require interpretation, coordination and judgment.

And the record itself becomes an increasingly valuable asset.

A complete, structured history of what happened, what changed, who decided what and why has implications well beyond efficiency. It matters to disputes. It matters to insurance. It matters to risk. And over time, it may change how firms think about the value of CA itself.

That’s a much more interesting future than simply doing today's work 10× faster.

The goal of AI in construction administration shouldn’t be to automate the architect out of the process.

It should be to automate the work that gets in the architect’s way.