I’m an Architect, What’s Your Superpower?

Here’s something I didn’t expect to happen: My architecture skills have become very relevant again.

For those who know me, you will know that I was a technical architect for many years. I actually taught enterprise architecture classes and wrote an architect certification program back when I worked at Microsoft. When I decided to move out of technical roles and become a Product Manager, I thought that part of my career was over.

But, I was wrong about that.

As a product manager, I found that having an architecture background was extremely helpful. It allowed me to ask questions about what we were building and (probably more importantly) why we were building it that way. Some engineering teams didn’t like explaining themselves to a PM but overall, I have had excellent relationships with my engineering peers and I think the fact that I have a technical background helps me work with that part of the product team.

Now that I work on AI toolchain, I find myself in a similar situation. As I’ve previously discussed, being a PM in the AI era is really great. You can hand specs to AI and get working code back. It’s truly liberating to build something that works by yourself.

However, when I talk to my students about their experiences, I find that they’re struggling. They get AI to build things and those things don’t work. When I ask them about the thing they built, I find that very basic architecture issues are holding them back. Web sites are built with all the code in a single large file. Or they build their own auth layer instead of just using something simple like Cognito. Or they don’t really have a backend and all the code is sitting there in the frontend. Etc. etc. etc.

For me, these things just make sense. If you are building a SaaS site (or anything, really) you want to ask the how and why questions. How is it going to be built and why is it going to be built that way?

The fascinating thing about AI is that you don’t really have to know that much about architecture. You need to know that architecture is a thing and you have to be curious about it, but you don’t have to be an expert.

Unlike very senior engineers that I’ve worked with, AI will patiently answer your questions about how the app works and why it works all day. Talk about “no stupid questions”, AI literally doesn’t care. It will answer your question earnestly.

Of course, current AI tools are not as good at architecture as a really good human engineer would be, but to be honest, that’s not really a huge issue. In the vast majority of cases, the software you are building isn’t that complicated. If you are building a SaaS site to help eBay sellers manage their online stores (this is an actual project from one of my students), I can assure you that even an “average” architect can design it for you just fine. And that’s well within the capability of AI tools like Claude.

Does this mean that you have to become an expert at architecture to build with AI tools?

No.

I am not an expert at architecture. I have some background in this area and I’m interested in it, but I haven’t been a practicing architect for almost twenty years. None of my technical knowledge from that part of my career is relevant to the projects I’m building now.

But I do know the questions to ask.

So, when you are building a project with your AI toolchain, start with Architecture. Ask the LLM to propose a detailed architecture plan. Read that document. Ask questions. Focus on the why part. Why did it do what it did. What are our alternatives? If the site grows to 100K customers, what happens? Can we build this site incrementally? What parts are mandatory before we launch an MVP and which parts can be put off to a future phase? Is this architecture conformant with industry best practices? Does this architecture plan ensure that the site will be secure and easy to maintain? How will hosting cost scale as we add users? What is the most expensive part of the planned architecture? How can we contain cost without sacrificing functionality?