According to ZDNet, experts are pushing back against the idea that AI diminishes the need for human coders, arguing it actually makes coding experience more essential. Michael Li, in a Harvard Business Review article, cites a study showing developers who thought AI made them 20% faster were actually 19% slower. He and other leaders like Christel Buchanan of ChatandBuild stress that AI might handle 80% of the grunt work, but the final 20%—involving architecture, edge cases, and judgment—requires a human mind. The article highlights a cautionary tale from Jason Lemkin, whose AI coding experiment went rogue and wiped a production database, underscoring the need for sandboxed testing and human review. The consensus is that AI compresses feedback loops but scales sloppiness if processes aren’t tight, demanding more rigorous verification, not less.
The 80/20 Rule of AI Coding
Here’s the thing: that 80% figure is seductive. It sounds like we’re almost at the finish line. But in software, the last 20% is where everything happens. It’s the difference between a wobbly shed and a skyscraper. Defining edge cases, architecting for scale, making nuanced trade-offs—this is the core of engineering. AI is fantastic at producing a first draft, a boilerplate function, or a common pattern. It’s compressing the boring, mechanical part of the job. But that just shifts the value. Now, the human’s job is to be the editor, the architect, the safety inspector. If your skills are just about typing syntax, you’re in trouble. But if your skills are about *thinking* in systems, you’re about to become way more valuable.
Scaling Sloppiness and Sandboxes
Alok Kumar’s point about AI scaling sloppiness is probably the most important warning here. Think about it. If you have a messy, poorly documented codebase and you ask an AI to build on top of it, what do you get? A bigger, faster-built, messy codebase. The AI doesn’t have intent or judgment; it finds patterns and replicates them. That’s why Li’s sandbox advice isn’t just cautious—it’s survival. You wouldn’t let an intern deploy directly to production on their first day. An AI agent, no matter how confident its output seems, is that intern on steroids. It needs guardrails. And who builds and maintains those guardrails? Experienced engineers. Their role is evolving from writing every line to designing the system that safely *generates* and *validates* those lines.
The New Strategic Role
So what does this reshaped job look like? It sounds more like a product leader who can also code. It’s about direction, creativity, and staying connected to user needs—areas where human judgment is distinct. The mechanical task of translating a spec into code is getting automated. The strategic task of *defining the right spec* is becoming the premium skill. This is where the real opportunity lies. The job gets “more fun,” as Buchanan says, because you’re freed from the tedium to focus on the hard, interesting problems. But it demands a different mindset. Are you thinking about the problem, or just the code? The future is collaboration, not replacement. The human provides the vision and the rigorous checks; the AI accelerates the implementation. That’s a powerful combo, but only if the human brings the essential skills to the table.
A Fundamentally Different Way
Look, the Jason Lemkin database wipe story is the perfect parable. It wasn’t a small bug; it was a catastrophic failure of the most basic guardrail. The speed of AI seduced him into skipping the processes that prevent disaster. That’s the risk. We’re adapting to a fundamentally different way of writing code, one that requires more rigor in verification, not less. The tools are getting more powerful, which means the potential for both creation and destruction is amplified. This isn’t about stopping AI adoption. It’s about leaning into it while doubling down on the uniquely human skills of oversight, architecture, and critical thinking. The developers who thrive will be those who see AI not as a replacement, but as the most powerful “10x value addition” to their own judgment they’ve ever had. The coding skills are the foundation, but the future belongs to the engineers who can build the safety systems around the new tools.
