How Software Engineering Evolves When Everything's Powered by AI

In the good ol' days, software engineering was all about building neat and tidy systems, each component neatly boxed and labeled. But then AI came along, kicked down the doors, and decided to rearrange the furniture. Now, every system worth its salt is AI-enabled, leveraging the power of LLMs (Large Language Models) to do the heavy lifting—whether they're public LLMs or those fancy ones hosted by cloud giants like AWS Bedrock.

So, what's it like to be a software engineer in this brave new world? Spoiler alert: It's a bit like trading in your reliable old bicycle for a shiny, self-driving car that occasionally decides to take a detour just for fun.

System Design Doesn’t Follow the Old Rules

Remember when you could design a system with a clear set of inputs and outputs? Well, with AI, it's like trying to predict the weather... in a sci-fi movie. LLMs are powerful, but they're also unpredictable. You give them input, they give you output, but what happens in between is a complex dance of probabilities, data patterns, and some AI magic.

Software engineers now need to account for this unpredictability. It's no longer just about writing code; it's about managing and mitigating the quirks of AI. You have to design systems that are flexible, resilient, and capable of handling the occasional oddball output without throwing a tantrum. The old adage "garbage in, garbage out" has turned into "interesting in, surprising out."

Reflecting on this evolution, I often find myself thinking back to the foundational principles of software engineering laid out by Dr. David Parnas. His iconic paper on modularity and information hiding revolutionized the field and became a bedrock of design methodology. During my Ph.D. journey, having Parnas on my thesis committee was a profound privilege. His teachings not only influenced my academic pursuits but also deeply shaped my approach to software engineering, especially as I navigate the complexities of AI-driven systems today. His work is a constant reminder that even in a world dominated by AI, the core principles of software construction remain essential.

Testing Is No Longer Black and White

Once upon a time, testing was straightforward. You wrote some tests, ran them, and if everything was green, you patted yourself on the back. But AI? Oh no, AI laughs in the face of binary testing.

With LLMs, testing becomes an exercise in managing uncertainty. Sure, you can check if the output is correct, but "correct" is often subjective. Is the response relevant? Is it coherent? Does it make you question your life choices? Testing AI systems requires a blend of automated tests and good old-fashioned human judgment, with a healthy dose of “it’s good enough” thrown in.

Monitoring Feels Like Herding Cats

Monitoring AI-driven applications is like trying to keep an eye on a herd of cats that sometimes morph into tigers. Abnormal usage patterns, sudden spikes in traffic, incorrect API usage—these aren't just metrics; they're your new best frenemies. And just when you think you've seen it all, you'll be hit with something new, like unauthorized data transfers or a spike in latency that makes you nostalgic for dial-up internet.

To manage this unpredictability, you can’t rely on AI systems to come with built-in tools. Instead, you need to set up comprehensive monitoring, manage sudden traffic changes, and keep a close eye on API usage to catch errors before they escalate. It’s also crucial to maintain AI project inventories, develop detailed risk registers, and implement robust audit and compliance reporting. These measures are essential components of a resilient AI system, turning you from a coder into a full-fledged risk manager and policy enforcer.

In summary, software engineering in the age of AI is like navigating uncharted waters with a compass that sometimes points wherever it pleases. The old maps don’t quite apply, and the landmarks keep shifting. Yet, this uncertainty brings unprecedented opportunities to innovate, adapt, and redefine what’s possible. If you’re someone who thrives on solving puzzles and doesn’t mind a few surprises, there’s no better time to dive into the ever-evolving world of AI-powered systems. Just be prepared—this journey is anything but predictable, but that’s exactly what makes it so rewarding.