If you’re not coming out from under a rock, you may have heard that it’s the end of the road if you’re making software today. AI is coming to take your job! We’re entering an age of robots building software! Get on board now, or your competitors will squash you like a bug! There’s a lot of hype … and do I smell tulip bulbs?
Frankly, I suspect that’s why we’ve seen a downtick in inquiries and guidance sessions about DevOps. Some enterprises are hoping they can get rid of those costly/scarce software people and replace them with subject matter experts and AI. If coders are gone, why invest in a system to control the software development lifecycle? I believe enterprises choosing that path are about to make a costly mistake.
But let’s just say they’re right and I’ve completely missed the train. Tomorrow’s vibe coders will still need a lot more discipline than they realize. In fact, they’ll need a DevOps platform or something very similar.
You Still Need To Control Source
Source code is the expression of business requirements codified. Let’s assume the AI writes everything for you perfectly, somehow guessing exactly what you want and never hallucinating or misinterpreting. At that point, will you need any source? Surprise! You’ve got source already. It’s just moved up one level of abstraction — you’ve been calling it prompts. You still need to store these as you make changes to the system, add new requirements, update old requirements, and fix the assumptions that you made that turned out not to be true. In fact, you may want to store more than before, since generating code with AI isn’t deterministic — odds are good you’ll need to revert to old working code more frequently. You need source code management, just like what you’ve got in your DevOps platform.
You Still Need To Build And Integrate
Shiny vibe-coded demos are great, but enterprise-class software is going to have more than one person working on it. That means collaboration and integration. Assuming all those predictions about productivity are right, it means a lot more integration. And subject matter experts won’t be experts in making their software work well with others. That means a build pipeline to automate builds, and you need to manage the change from several experts. You need the continuous integration that you’ve got in your DevOps platform.
You Still Need To Test
Once you have something that’s executable, you need to prove it works. AI introduces a host of exciting new wrinkles into your life. You don’t just have to make sure your chatbot works and answers correctly when prompted. You also have to make sure that it doesn’t introduce bias or start offering sweetheart deals. That requires testing, and that testing can’t all be done by hand. Once again, the name is different — the cool kids call them “evals” — but it’s really automated testing. You need continuous automated testing, just like what you’ve got in — or integrated with — your DevOps platform.
You Still Need To Secure
AI has opened brand-new opportunities for malicious actors. Regardless of how you build your code, you need to make sure that it’s not subject to prompt injections or jailbreaking. Beyond that, underneath everything is still just code. How can you make sure that AI hasn’t added vulnerabilities? You’ve got to scan the generated code and keep run-of-the-mill SQL injections and cross-site scripting out of your application. And you want to keep a close eye on your models — especially if data scientists are tweaking them. You need security scans and software supply chain controls, just like what gets run by your DevOps platform.
You Still Need To Deploy
As one vibe coder learned recently, it’s a bad idea to give your AI unfettered access to prod. You don’t want to discover when your hosting bill arrives that AI has decided the best platform for your documentation is an AWS 16xlarge high-performance compute server. You want your deployment process to be deterministic, repeatable, routine, and — most importantly — dull. Excitement is for users. When it comes to getting bits on servers, you want the same thing to happen every time, with predictable costs. You might use AI to generate infrastructure as code, but once you’ve done that, you’ll want it locked down and in budget. New features still need to trickle out gradually so you can see how users react. You’ll need the same deployment technology that you’ve already got in your DevOps platform.
Tomorrow’s AI-Enhanced Developers Need Today’s Practices
In short, the AI developers of tomorrow will need strong grounding in the basics of the software development and delivery lifecycle. They’ll need to think about building software the same way we do today. Every article I read about vibe coders losing all their work due to a prompt that went awry, or releasing an app that gets exploited on day one, or building a chat app that offers a car for $1, no takesies-backsies — they all reinforce my belief that AI is a compelling tool but only one tool in our toolbox. We’ve built a great body of practice and powerful platforms to help people evade many of the dangers in software development. It’s painful to watch vibe coders rediscover the need for them, one avoidable fiasco at a time.
