Building Software in the Age of AI: Speed, Leverage, and Lost Context

‍

It’s been nearly three years since ChatGPT launched. In that time, I’ve used it – and many other AI tools – almost daily, sometimes as a builder, sometimes as a manager. I don’t have to tell you it’s been a game changer. AI has completely transformed the how of software development. But it’s also changing what developers and teams need to care about when they build software.

‍

The Upside: More Speed, More Reach

‍

The most obvious benefit of coding with AI is speed. You can go from idea to prototype in minutes, and to a functional application shortly after. AI tools handle boilerplate setup, repetitive code, and even documentation – freeing you up to focus on higher-level design and strategy. The feedback loop between idea and implementation has never been tighter, especially at the start of a project.

‍

AI also solves one of the hardest parts of any creative process: starting from a blank page. It gets you from zero to one. Whether you’re technical or not, you can take a rough concept and quickly spin up something tangible. That initial momentum helps you test and iterate fast – whereas without AI, you might never get past the starting line.

‍

The Downside: Shallow Understanding, Shaky Learning

‍

For all its upside, AI hasn’t fully changed the game – especially at the team and organizational levels. The same acceleration that helps you build faster can also make it harder to understand why something works. You can build an entire app without ever learning the framework behind it. That’s fine in the short term, but over time it creates fragility. Debugging takes longer, tech debt piles up, and the team’s overall competence can stagnate.

‍

As a manager, I’ve seen this play out with junior developers. AI gives them a boost – but also a ceiling. When you can instantly generate working code, you’re less likely to read the docs, explore the framework, or understand design trade-offs. You might even deploy code you don’t fully grasp. AI can fix the immediate issue, but it often creates new ones down the road.

‍

The Context Problem

‍

The other major limitation is context. AI coding tools operate in short bursts – they can see your current file, maybe a few hundred lines around it, but not the full system. That means they can suggest code that looks fine in isolation yet breaks patterns, violates performance constraints, or misaligns with business logic. And after 10 or 20 iterations, most tools start forgetting the original intent entirely.

‍

Until AI can maintain a persistent, holistic understanding of a codebase, human oversight isn’t optional – it’s essential.

‍

A Smarter Way Forward: AI + a Human in the Loop

‍

AI tools aren’t going anywhere. They’ve already transformed what individuals and small teams can achieve. But both AI and humans have blind spots – and ignoring them is costly. Used well, AI is an accelerant; it provides leverage. Used carelessly, it erodes context and craftsmanship.

‍

The best path forward isn’t AI or human – it’s both, together.

‍

That belief is what led us to build ModernSchema. We saw AI and no-code tools exploding in popularity but also witnessed their limits firsthand – especially when it came to production-grade software. Our approach – AI + a human in the loop – offers the speed and leverage of AI while preserving the oversight and context needed to ship production-ready code.