The best age for engineers is here

The question presents itself with the clarity of a mathematical formula: if artificial intelligence can code as well as humans by 2026, as Anthropic's CEO Dario Amodei predicts, what remains for the human engineer? The answer is nothing less than the freedom we have sought all along.

I recall my own first encounter with programming in high school, when Visual Basic was the new kid on the block. That first program was like finding a hidden door in your childhood home. Suddenly the world opened up—anything seemed possible. Freedom, that's what it was. Pure freedom.

Yet in the twenty years since, that initial exhilaration has been tempered by reality: ideas accumulate faster than the hours needed to implement them. Days, weeks, months are spent on the grunt work - type, debug, refactor, repeat.

As AI coding tools emerge, I see a genuine opportunity to change how software is built. I evaluated them as new ones came out, and I've established a consistent benchmark: creating a web crawler that extracts meaningful links while filtering out noise. It's a practical problem with clear success criteria. My methodology is strict: I don't write or even look at the code, evaluating only the final results.

Claude Code

Claude Code was the latest tool I tested and the first to truly pass my evaluation.

First Attempt: The practical test

Claude Code operates through a straightforward command-line interface requiring terminal and filesystem access. After granting these permissions, I provided a brief description of requirements and it went off to code.

What impressed me immediately:

1. Intuitive interface: Virtually no learning curve or setup as it runs directly in your terminal.
2. Environment management: Fast to hit the ground running. It handles project setup, dependencies, and configuration.
3. Self-improving cycles: It creates tests, identifies failures, and fixes bugs autonomously.
4. Human-like methodology: It uses pragmatic approaches: grep code snippets, testing incrementally, and iterating.
5. It compromises! Unlike other tools that get stuck in perfection loops, Claude recognizes when minor issues can be deferred.

Areas needing improvement:

• Context window limitations: Performance degrades as the context grows, making it less effective for complex, long-running projects.
• Permission management: It requests approval for each new command, which becomes tedious. A true "YOLO mode" where it operates autonomously within defined boundaries would improve workflow efficiency.
• Algorithm interdependencies: The system struggles when multiple algorithms need coordinated changes to work together effectively.

The system's first attempt produced a functional crawler but with notable limitations:

• Too many irrelevant links included
• Vulnerability to network errors
• Inconsistent crawling depth behavior

As I provided additional requirements, I observed a familiar pattern: increased complexity led to decreased quality. The system struggled to maintain cohesion as more interconnected logic accumulated across context windows.

Second Attempt: Requirements-driven development

When requirements are given in multiple prompts, Claude has a hard time to keep up, especially when context is reset due to its growing size. So I started from a fresh new folder, and instead of incremental guidance, I created a comprehensive requirements document specifying:

1. Functional requirements
2. Quality attributes
3. Implementation guidance

The document was concise, just one page of markdown, but comprehensive enough to provide a global view for Claude to reference. I launched it with a simple prompt: "read requirements.md and go."

The results were remarkable:

• Modular architecture with well-defined components
• Comprehensive test cases for each module
• Behavior closely matching requirements
• Few logical or implementation errors

Testing against 50+ complex websites showed excellent performance. The entire process took approximately 1 hour instead of a day or two.

The crawler project is open-sourced here for anyone to use.

The best age for engineers isn't behind us. It's here, now.

We are not witnessing the twilight of the engineer but rather the dawn of their golden age—a time when human ingenuity, freed from implementation constraints, can focus exclusively on the problems worth solving and the ideas worth pursuing.

And perhaps most importantly, this newfound freedom grants us the luxury that has always driven innovation—the ability to entertain silly ideas and bring them to life.