AI coding tools such as Claude Code are rapidly changing how software engineers work, and Obie Fernandez, chief technology officer at Zardot, says the shift is already large enough to threaten a major share of developer jobs.
In a recent essay titled You’re Already a Meat Proxy, Fernandez wrote that more than 80% of people working in software development may already be facing career obsolescence, even if many of them do not realize it yet. Fernandez has more than 30 years of software development experience. He previously served as a principal engineer at Shopify, chief technology officer at Andela, and is also a partner at MagmaLabs and an author of technical books.
A friend was already using AI, but still could not move work forward
Fernandez said the argument sharpened after meeting a friend who works as a Staff Engineer at a large and successful startup with hundreds of engineers. According to Fernandez, that friend had spent six months complaining that he no longer knew how to keep doing his job.
When Fernandez asked him to open his work laptop, he saw dozens of browser tabs and several terminal windows already running Claude. Fernandez then asked a direct question: if AI tools were already in use, why was no real progress being made? The engineer replied that he had tried to use Claude to solve problems, but had not been able to get anywhere meaningful.
Fernandez took over, switched the model to Fable, which he described as more suitable for coordinating development work, and entered a very simple prompt despite having no familiarity with the project architecture: 「This code is bad, and the whole architecture is a mess.」 The friend laughed and questioned what such a vague instruction could accomplish.
Fernandez said the AI immediately began scanning the repository and surfaced a long list of technical debt issues, including inadequate test coverage, giant objects with thousands of lines of code, and dead code that no longer served a purpose. After a few confirmations, the system deleted part of the unused code and produced a draft pull request.
That was where the process stopped. The engineer asked Fernandez to halt, saying he could not simply change company code on his own and that all changes had to go through team discussion, planning, and approval. Fernandez responded that Claude could write the plan as well. After confirming that Claude was connected to Jira through MCP, he asked the AI to convert the proposed changes into Epics and Stories. The system created five Epics and was preparing to generate related work items when the engineer pressed ESC to cancel it.
Fernandez said the obstacle did not appear to be technical competence. In his view, the real issue was that the engineer could not accept that AI was changing his mode of work. He quoted the friend as saying, 「But I want to use my own brain!」
"From L1 to L7, everyone is doing the same thing: talking to Claude"
Fernandez also cited another post that circulated among engineers. X user @v0xium wrote that after only half a month at a large company, he had become disappointed with the AI-driven software development model he saw there.
According to that account, nearly everything on the team was being generated by Claude Code: requirement specifications, code, tests, product requirement documents, Jira tickets, and reports. Management had not reduced the workload. Instead, it pushed engineers to deliver faster, and some employees were working 12 to 13 hours a day, with much of that time spent checking AI output.
The engineer wrote, 「No one is actually reading anything. From L1 to L7 engineers, everyone is doing the same thing: talking to Claude.」 He said the team was losing technical understanding and the sense of accomplishment that comes from solving problems directly.
Fernandez drew a different conclusion from that complaint. If an engineer’s value is mainly to receive tasks, follow a preset process, and pass the result to the next person, he argued, that role was always vulnerable to automation.
What Fernandez means by "meat proxy"
Fernandez uses the term "meat proxy" for workers whose core function is to translate and relay instructions inside an organization. That could mean an engineer turning a manager’s request into code, or a mid-level manager converting executive direction into tasks for subordinates.
He said those roles held clear value in traditional software development because translating business requirements into working software required large amounts of labor and technical skill. But once AI can understand requirements, write code, generate tests, organize documentation, and even operate Jira, the pure instruction-translation layer loses much of its original economic value.
Fernandez added that many engineers still have jobs today largely because of organizational inertia, with efficiency gains from AI not yet fully reflected in staffing. He wrote, 「Current AI technology is already enough to replace all meat proxies, but disruption at this scale takes time.」 Based on that view, he said more than 80% of software development workers may already be in career danger.
He says autonomous AI agents now handle 75% to 80% of his coding work
Fernandez also described his own workflow to show how AI is changing senior engineering productivity. He said he is primarily using Ruby on Rails to build a new internal enterprise system. Because the architecture is sound and the test coverage is strong, he does not feel the need to inspect every line of AI-generated code.
Over the past few months, he said, he has stopped directly supervising Claude Code line by line and instead allows autonomous AI agents to carry roughly 75% to 80% of his coding work. He steps deeper into implementation only when adding major new features or making large architectural changes.
That shift, he said, lets him spend more time on decisions he sees as more valuable: understanding user needs, deciding what should be built, and determining how a system can produce real business value.
He also acknowledged that his own environment differs from his friend’s. The friend works on a live production system tied to customer payments, where any change can create real risk. In that setting, line-by-line review of AI modifications may still be necessary.
Moving upward in the org chart may not protect engineers
Fernandez pushed back on a common suggestion that engineers should retrain as software architects because AI will always need humans to handle high-level design. He said one of his own main sources of professional value over the past 20-plus years has been system architecture and engineering judgment, yet he expects AI to gradually take over more of that work as well.
He pointed to tasks such as clarifying requirements, identifying system boundaries, setting testing standards, and judging whether a new architecture satisfies obligations created by an existing system. In his view, those capabilities are not guaranteed to remain exclusively human. As he put it, 「Moving one box higher on the org chart does not mean you are outside this process.」
For Fernandez, the real distinction is not job title. It is whether a person is willing to identify problems on their own, make judgments, and take responsibility for outcomes. Even then, he said, no one can guarantee they have abilities that AI will never replace, including himself.
More code output does not automatically mean more value
Fernandez said some people simply want a stable, high-paying job that leaves room for family, travel, or life outside work. He said there is nothing wrong with that, but questioned whether the software industry will continue to offer that model to everyone in the future.
For those who want to preserve their current lifestyle, his advice was to use AI to meet work demands and spend the extra time on family and other options in life. For engineers who want to create more value in the AI era, he said the better path is to look for problems worth solving, speak with people directly affected by those problems, and then use AI to research, challenge assumptions, and execute.
He ended by reflecting on the session with his friend. In about 20 minutes, Claude had produced a draft pull request and several Jira Epics. That still did not prove the changes improved anything. Fast code generation, he wrote, is not the same thing as creating value. The key question remains which problems are worth solving and whether the proposed solution actually works.
Fernandez closed with a direct line: 「You do not become less of a meat proxy just because you personally typed a few more lines of code.」 As AI capability continues to rise, he argued, engineers may have to think less about preserving old ways of working and more about what they actually want, and how to use new tools to get there. He ended with another challenge: 「You really want to keep using your own brain? You do not want your brain to atrophy? Then figure out what you actually want, and let Claude work for you.」