The faster AI writes code, the more architecture matters.
The strange thing about AI coding is that I am writing less code, but thinking more about software.
I notice this more and more in my own work.
Agents can build surprisingly fast now. I can give them a direction, let them work, review what comes back and keep moving. Things that used to take hours of coding can sometimes be done while I am busy with something completely different.
But that doesn't mean I have less work to do as an architect.
Actually, I find myself thinking much more about what surrounds the code.
Is the architecture still clean when the system grows?
What did the agent assume that I never explicitly told it?
Where are the security boundaries?
What happens when something fails?
Can I trust this in production, or does it just look like it works?
And perhaps most importantly, did we build the right thing in the first place?
This is where I think the conversation about AI and software development sometimes becomes too focused on productivity.
Yes, we can produce code much faster. I see that myself.
But producing code was never the hardest part of building serious software.
The difficult part is making hundreds of small decisions that eventually determine whether a system is solid, secure, maintainable and actually useful.
AI doesn't make those decisions disappear.
If anything, when execution becomes this fast, poor decisions can become code even faster.
After almost 30 years in software, I find that shift fascinating.
I spend less time telling a computer exactly how to build something.
And more time thinking very carefully about what I am asking it to build, why, and what I am willing to trust.
Maybe that is where the role of the software architect is heading.
Less code. More judgment.