There is a pattern I repeat whenever someone asks me if AI is going to steal dev jobs:
Excel did not kill the accountant. It killed the spreadsheet filler.
In 1985, the worry was legitimate. Hundreds of professionals spent hours moving numbers from one place to another, adding up columns, cross-referencing tables by hand. Excel automated exactly that.
What happened: the profession did not disappear. It transformed. The ones who survived were the ones who knew what to do with the numbers after they were calculated. The ones who understood business, taxes, risk. The spreadsheet filler went away. The accountant stayed.
I am watching the same movement happen now, at a faster pace.
The split is in the data
Boris Cherny, creator of Claude Code, said in January 2026 that the title “software engineer” will start to disappear. Not because nobody will be building software anymore. Because the people building it will call themselves something else.
What I see in the market points in the same direction: job postings have been pulling toward profiles with architecture and systems thinking, while pure implementation loses ground.
GitHub Copilot already generates about 46% of the code written by active users of the tool, according to GitHub itself. Claude Code and its competitors are completing entire blocks of code from a description. Productivity went up. The number of people needed to produce the same amount of code went down.
The coder, who spends 80% of the time writing code that follows a pattern, is being replaced. Not by an AI on its own. By an engineer using AI.
The anti-pattern I see in companies
The wrong answer is hiring a “jr prompt engineer” to replace a jr developer.
I saw this at two Brazilian companies in 2025. The logic was: if the AI writes the code, we need someone who knows how to write prompts. The result: code quality plummeted, integration bugs exploded, and the team took 3x longer to debug because nobody understood what the code was doing.
Swapping a coder for a coder 2.0 changes nothing structural. AI produces code. But it does not understand business context, it does not anticipate how that service will interact with Itaú’s billing system, it does not know that endpoint will get called ten thousand times per second on a Pix rush the night before payday.
That is engineering work. And it was never a coder’s job.
What the market will pay for over the next 3 years
The coder translates spec into code. The engineer understands the problem before translating, questions the spec when it is wrong, and architects for what comes next.
The coder executes tickets. The engineer sees the whole system and knows where that ticket will create a problem 6 months from now.
The coder uses AI as souped-up autocomplete. The engineer uses AI as a codesigner and spends 70% of the time reviewing and validating what the AI produced.
The logic is the same as any automation. The work that remains is not writing more code. It is making sure the code that exists does what it needs to do, at the scale it needs, with the security the regulator will demand.
I do not know what my job title will be in 2028. But I know the work will be the same as it has always been: understand the problem, design the solution, make sure it works in the real world.
AI will take care of the template code. The rest is still the job of whoever thinks.