Anand Naidu, our resident development expert, has built a career navigating the frontiers of both frontend and backend engineering. With deep proficiency across multiple coding languages, he offers a grounded perspective on the seismic shifts AI is causing in the software world. Today, we delve into the practical realities behind the AI hype, exploring the current friction between AI’s potential and its performance, the enduring value of human craftsmanship in coding, and how business leaders can strategically embrace this new era without getting burned. We’ll discuss why some AI features fail while others succeed, and what the future may hold for the very role of the programmer.
Some view current AI coding assistants as a “flickering light bulb”—brilliant one moment, then dark the next. Considering this, what specific tasks or qualities make a human junior programmer more reliable today? Please provide an example of a coding challenge where AI typically fails but a junior developer would succeed.
That “flickering light bulb” analogy is perfect because it captures the core issue: inconsistency. You can be in total darkness, then suddenly the AI flashes on and you think, “Wow, I can see everything!” But two seconds later, you’re plunged back into pitch black. A junior programmer, even while learning, provides a steady, consistent light. Their reliability comes from their ability to reason, to debug methodically, and to understand the broader context of a project. For example, imagine a complex bug that’s buried deep within a legacy system. An AI might generate a clever-looking snippet of code that addresses a surface-level symptom, but it often misses the nuanced, interconnected logic. A junior developer, on the other hand, can be guided to trace the problem step-by-step, to understand the ‘why’ behind the code, and to implement a fix that respects the entire system’s architecture. Frankly, the output from many AI tools today simply isn’t as good as the work I’ve seen from most junior programmers I’ve had to work with.
For 37 Signals’ product Fizzy, humans reportedly coded 95% of it, and potential AI-powered features were shelved. Could you walk through the evaluation process for those AI features? What specific performance gaps or user experience issues led to the decision to place them on the cutting room floor?
When you’re building a product like Fizzy, user trust and reliability are everything. While we were certainly excited and experimented with AI-powered features, they ultimately ended up on the cutting room floor because they failed the consistency test. The evaluation process is deeply practical. We ask: Does this feature consistently deliver value, or does it create unpredictable and frustrating experiences? The issue wasn’t that the AI was never helpful; it was that its performance was erratic. One moment it might generate a brilliant suggestion, but the next it would produce something illogical or unhelpful. For a user, that’s a jarring experience. It breaks their workflow and erodes their confidence in the tool. We’re not seeing any evidence that we are falling behind by prioritizing this human-centric approach; in fact, our ability to produce and launch quality products remains as strong as ever.
The idea of hand-coding being a luxury craft, akin to modern saddle making, is a powerful image. How does this philosophy shape team culture and development practices at 37 Signals? Looking forward, do you see this “artisanal” approach remaining viable, or is it an indulgence of a passing era?
That philosophy is central to our culture. It fosters a deep sense of pride and ownership in our work. Like a master saddle maker who appreciates getting the leather and the stitching just right, we care about the beauty and integrity of the code we write. It’s a luxury, absolutely, to be able to focus on that level of craftsmanship in today’s environment. This shapes our practices by encouraging us to build things that are not just functional, but also elegant, maintainable, and a joy to work on. As for its viability, I’m not concerned. We might not be part of the main production line for “transportation”—the mass-produced, AI-generated code—but so what? There will always be a market for high-quality, handcrafted goods. I’ll continue to make my “handwritten code saddles” for as long as I can, simply for the enjoyment and the superior result it produces.
Some AI agents, like Shopify’s SiteKick, are delivering tangible benefits for merchants. What distinguishes a successful, high-value AI implementation like this from the less practical applications in software development today? Please detail the key factors that separate a game-changing AI tool from a novelty.
The distinction is incredibly clear: successful AI tools provide undisputable, tangible benefits to a specific user problem. Look at what Shopify has done with SiteKick. It’s an AI agent that helps merchants set up and optimize their shop. The value is immediate and measurable for the user—they build a better store, faster. That’s truly incredible. It’s not an abstract tool for a developer; it’s a solution for a merchant. The less practical applications are those trying to replace the core, nuanced work of development, where that “flickering bulb” problem becomes a fatal flaw. A game-changing AI tool is one that is focused, reliable within its domain, and directly empowers the end-user to achieve a goal. A novelty, on the other hand, is a general-purpose tool that demonstrates a capability but lacks the consistency and focus to be truly integrated into a critical workflow.
The U.S. economy is making a massive bet that AI will eventually overcome its current limitations. Given this high-stakes environment and the uncertainty you described, how should a standard SaaS business navigate the hype? What practical steps should leaders take to prepare without over-investing in immature technology?
It’s astonishing to witness. The entire American economy feels like one colossal bet that we’ll reach something like Artificial General Intelligence, and the conviction is so strong that the investment is essentially limitless. For a standard SaaS business, jumping all-in on that bet right now would be reckless. My skepticism isn’t about the long-term potential, but about the current reality. The claim that you can fire half your programmers and go faster is simply not what we’re seeing. The best approach is to accept the uncertainty and stay agile. The world is going to look totally different, much like it did after the iPhone’s release in 2007. The wisest thing to do is to just try to hold onto your hat. Be the cowboy who jumps on for the ride to see where it goes. Practically, this means you should experiment, stay informed, and build a culture that can adapt, but don’t gut your most reliable asset—your human development team—based on a promise that hasn’t yet been fulfilled.
What is your forecast for the role of programmers over the next five years?
My forecast is one of evolution, not extinction. You simply don’t get to roll these things back; this technology isn’t going away. Over the next five years, the role of the programmer will become even more focused on architecture, critical thinking, and quality control. As long as AI remains a tool with brilliant but inconsistent flashes of insight, the programmer will be the steady hand that guides, validates, and integrates that output into a coherent, reliable system. The “artisanal” programmer who can craft beautiful, robust code will be in high demand to fix and refine what the AI produces. We are entering an exciting new era, and while we don’t know the final result, the need for human ingenuity to steer the ship has never been more apparent. Programmers will have to become masters of a new toolkit, but they will remain the essential architects of the digital world.
