Since 2025, AI assistants have become part of every developer’s daily routine. A year later, they’re indispensable. Those who ignore that will have even more catching up to do later. But between what’s being claimed and what’s actually possible, there’s a wide gap.
The Hype and the Reality
CEOs of Anthropic and OpenAI are announcing the end of software development, and the media are happy to repeat it: “In two weeks, a team of 16 AI agents replicated a C compiler that thousands of engineers spent 37 years building” — “A Google engineer reports that Claude Code rebuilt in one hour what her team had worked on for a year.” But look more closely, and you’ll see these are marketing headlines. This creates a distorted picture that many people base their decisions on.
A coding agent is only as good as the problem definition that precedes it. The AI compiler could draw on 37 years of documented experience and was ultimately just a slower, more error-prone copy. The Google project that was supposedly written in an hour was built on a year’s worth of selected ideas from prototyping and testing. These are examples of AI’s impressive ability to understand language and execute instructions. But what remains is what software development has always been about: understanding the problem.
AI Only Accelerates What Already Works
Solving a problem doesn’t start with code — it starts with defining it. That happens through discussion and the exchange of experience. Anyone who builds software knows that most technical decisions are a compromise between several options, each with their own trade-offs. Over the course of a project, many such decisions get made, and often questions arise during development that no one thought of at the beginning. AI can only make those decisions if it has the necessary information. With the right tools, AI can gather information and weigh pros and cons much like a human would, to arrive at the best possible decision. But often, exactly that information simply isn’t available.
Why Vision Remains Human
A good product doesn’t win people over through technical excellence. Marketing contributes a large part to a product’s success — and that requires creativity, something AI still cannot reproduce. Decisions are emotional, and a human’s capacity to absorb and synthesize information exceeds what can be fed to an AI.
A second critical factor is the onboarding experience: how quickly does a new customer understand that a product can help them? A perfect example is ChatGPT: a text field where you can type anything you want. Here too, AI can’t help reduce the user experience to its essentials — it lacks the vision to do so. Finally, it’s about binding customers to your product through features that complement each other without undermining the original experience. Code is written to leave room for extensions that don’t yet exist. This confronts developers with constantly shifting architectural requirements — a challenge where experience plays a major role.
What This Means for Software Development
Software development will increasingly move away from detail questions like syntax and API integrations. AI can already handle those tasks very well. What will remain necessary is the interface between humans and machines — someone who understands human requirements and translates them into instructions for AI. If you need a developer who understands both the technical aspects and your needs, I’m happy to help.