W30 - Full-Stack Development in the AI Wave

Maintain cautious optimism about full-stack.

Full-stack has existed for a long time — individual developers can often be considered full-stack — but it never became mainstream domestically. Large companies tend to favor deep technical expertise, and interviewers usually dismiss jack-of-all-trades full-stack candidates. Now the value of full-stack is being re-evaluated; I think this shift calls for a cautiously optimistic attitude.

Many of the generative AI–related technologies in this wave are near the top of the Hype Cycle, including Vibe Coding and Software Engineering. Technologies at their peak are often accompanied by excessive expectations and impractical deployment behavior.

Vibe Coding and Software Engineering are different. Vibe Coding: AI is good at producing one-off prototype code, quickly assembling runnable products to validate designs and user acceptance. Because these are prototypes, concerns about later maintenance and code quality don’t apply. Software Engineering: coding itself is only part of routine delivery work. Research from organizations like METR shows that AI tools provide very limited productivity gains for highly experienced developers.

Mastering fundamentals lets you accomplish 99% of tasks; fundamentals remain important. While troubleshooting an issue last week, I clearly felt gaps in people’s knowledge of basic networking (like Nginx and the HTTP protocol) and company middleware (PaaS layer). When systems run normally everyone thinks it’s simple, but when problems occur they fall into awkward blind searches. If the fundamentals are weak, even the most advanced intelligent tools can’t create reliable systems.

Engineers need time to build trust with AI and develop an accurate sense of its capability boundaries. There are three boundaries to explore in practice: when you can trust AI and involve it in tasks; when AI should produce the initial output and engineers serve only as reviewers; and when you must switch off AI and rely on people to resolve the problem thoughtfully.

Of course these boundaries are changing rapidly and intelligence levels are still improving, so be optimistic about AI’s positive impact on business development over the next 3–5 years. Last week OpenAI announced that one of its unpublished LLMs reached gold-medal level at the IMO (International Mathematical Olympiad) for the first time. Previously, only a Google DeepMind model optimized specifically for math had reached silver-medal level.