Building a SaaS Platform in the Age of AI

It's the biggest debate in technology, if you don't count when AI is going to bring about the end of civilization. What's the best way to use AI for developing products? Should you use it cautiously, or spin up agents upon agents to handle the designing, backlog creation, coding, and testing, with only human-in-the-loop gateways? Even as I type this, AI is trying to give me hints on what to type next. Stand down, AI — I've got this.

As this debate borders on nerd religion, I'm not here to argue there's a right way or a wrong way. Tech products are popping up everywhere, in every variety. I don't know that anyone has polled recent successful startups on how they used AI, and I'm certainly not volunteering. Our goal wasn't to find the perfect way to use AI — we just wanted to find the Headjacks way. We thought hard about where we wanted AI's help, and where we didn't.

Where We Did Want AI's Help

Syntax - I don't want to learn new languages at the syntax level anymore. I've learned dozens over my career, and guess what? Most are now obsolete, unsupported, or both. If I know I need a do-while loop, that's good enough — I don't need to remember the exact syntax. I've got more important things to remember, like where I left my car keys.

Root Cause Analysis - If something doesn't work as expected, I don't want to go searching for it myself. If you know how your product is supposed to work, it should be easy to describe what you expected versus what you got. We're just on our Beta 1.0, and our portal alone has code in 140 files and just under 25,000 lines, growing daily. Just locating the right file can take a while — AI is way faster at that than I am.

DevOps/Deployment/Initial Testing - I could give the simple answer and say I don't enjoy this part of developing software. Actually, that is the answer. And AI just does it better — don't fight me on this. Well, except for that one time a prod rollout sat there spinning for fifteen minutes and no AI flagged it for me. Someone still has to notice when something's off.

Functional Bug Fixes - I wouldn't say this applies to all bugs, but more to things that aren't working as designed, especially ones that throw a captured error. When we find bugs, we create Issues in GitHub, then point our AI tools at them. It's good at classifying them, prioritizing which to address first, and then systematically resolving, testing, and pushing fixes back to GitHub — which falls right into our DevOps process. Though I'll admit I had to point it at a report's existing working code before it found the real fix on one. A poorly designed feature that just isn't driving an outcome falls into the next category.

Where We Took the Wheel as Humans

Data Architecture and Movement - This isn't to say AI is bad at designing databases. As an analytics consultant myself, I've seen AI be pretty effective at it. But as architects at Headjacks, we wanted to be deliberate about our data, how it interacts with other data, and how it moves based on user activity. This is the heart of your platform. If you let AI design this in a black box, it's hard to answer basic technical questions or describe what actually happens when a user takes an action. I don't want to have to consult AI every time a customer asks why they're seeing what they're seeing. And for those root causes AI does struggle with, you can give it far better guidance when you not only know how something's supposed to work, but designed it that way yourself.

Your Overall User Experience - Sure, AI can execute your plan — but it needs to be your plan, described in enough detail for it to run with. Have you ever noticed how easy it is to spot when AI wrote something, like in a presentation you built with its help? AI can play devil's advocate on the experience you have in mind, but if there's one thing humans still do better, it's designing with empathy. Humans still know humans best, in most cases.

Your Ultimate Vision - You may have a good idea — something you wished existed in a moment, searched the marketplace for, and couldn't find, so you said to yourself, "that should be a product." AI can take that idea and run with it, and probably hand something back. But you can't make something game-changing if it's not something you're passionate about. You need extensive experience with the problem — cycle after cycle of thinking through every possible angle. This is a good place to put AI in the second chair. It can be great to talk something out with, but if it were that easy to solve, you probably wouldn't have a big market for solving it. And you won't have the passion to ride out the ups and downs of entrepreneurship, or the humility to admit when you got something wrong. Failing is the key to success!

In Reality

This is in no way an exhaustive list — it only scratches the surface. I didn't even get into important issues like cost or maintaining skillsets. These are just examples of where we put real thought into using AI, and where we chose to rely on ourselves first. There isn't a perfect way to develop with AI, and if there were, AI would evolve and change the game all over again. Headjacks was a deeply personal idea for us. AI has had a lot to offer and has probably increased our velocity 300%-500%, savings we intend to pass on to our customers. But at the end of the day, this product is a reflection of us. AI helps us get there faster, and makes our vision achievable with speed and quality.

Full confession: AI didn't just proofread this article, but actually pushed back on some of my points and helped me find more accurate ways to say them, even remembering good examples of how we have worked together. If AI is really taking over, I probably should stay on its good side and not steal credit for its work. Like a spouse, it has a better memory than you think!