The laziest narrative in tech in 2025 is this: "AI will replace developers." It's a statement that simultaneously overestimates AI and underestimates real software engineering work. The truth is more surgical. And for many people, more uncomfortable.
What AI Actually Does Well
Generative AI is extraordinary at pattern matching and code generation within well-defined domains. CRUD? Perfect. Authentication boilerplate? Instant. API endpoint that follows an established pattern? Trivial. And here's the problem: a huge slice of the software industry spent the last decade being paid to do exactly that.
What AI Cannot Do
There's a class of work that AI cannot execute — and may not be able to for a long time. It's the work of architectural judgment. Not just writing code — knowing what not to build, identifying where complexity will explode 18 months from now, and understanding the trade-off between consistency and availability when R$1 billion is in transit. No language model makes that decision for you — because it has no universally right answer, only a right answer for your specific context.
- Deciding what not to build
- Identifying where complexity will explode in 18 months
- Decomposing a business domain into services that make sense to the team, not just to the compiler
- Navigating consistency vs availability trade-offs in critical financial systems
The Assembly Line Analogy
When Ford automated the assembly line, carriage craftsmen panicked. What happened? The mediocre ones lost their jobs. The excellent ones created the automotive customization, performance and design industries. AI will do the same to software. The engineer who writes repetitive code without understanding the domain will be replaced. The one who uses AI to multiply delivery capacity while directing what to build will be worth 10x more.
AI & Engineering — test your understanding
What does generative AI do extraordinarily well?
Clique para ver a respostaPattern matching and code generation in well-defined domains: CRUD, boilerplate, authentication, standard endpoints. Everything that's repetitive and predictable.
What is "architectural judgment" and why doesn't AI have it?
Clique para ver a respostaIt's the ability to make technical decisions with context-specific consequences — what not to build, how to decompose domains, what trade-offs to make. There's no universal answer, only the right answer for that specific context.
What should you study to be irrelevant in 5 years?
Clique para ver a respostaSyntax only. Frameworks without understanding principles. Tools without understanding the problems they solve. Copy-pasting solutions without understanding the domain.
What should you study to be more valuable in 5 years?
Clique para ver a respostaDistributed systems and their trade-offs. Specific business domains. Communication with non-technical stakeholders. Decomposing complex problems into simple ones.
Stop asking "will AI replace me?". Ask instead: "how do I use AI to do in 1 hour what used to take a week, and direct my energy toward what it can't do?" That question has a much more interesting answer.
If you're afraid of AI, you're competing at the wrong level.
— Victor Fornitani