A tweet from AI researcher and entrepreneur Rohan Paul has gone viral, succinctly capturing a central anxiety in the tech industry: the shifting target of AI automation.
What Happened
On X (formerly Twitter), Rohan Paul quoted and amplified his own statement: “We’ve been telling kids for 15 years to learn to code. Well, AI is coming for the coders. It’s not coming for the welder…”
The tweet, devoid of additional commentary, served as a stark, provocative thesis. It directly challenges a long-standing career directive in education and tech—that coding is a future-proof skill—by suggesting the automation wave is now hitting the very field it was supposed to secure.
Context: A Shifting Narrative
The sentiment echoes a growing discussion among technologists and economists. For over a decade, "learn to code" has been a mantra for career resilience. However, the rapid advancement of AI coding assistants like GitHub Copilot, Amazon CodeWhisperer, and state-of-the-art models like DeepSeek-Coder and Claude 3.5 Sonnet—which can generate, debug, and explain complex code—has fundamentally altered the calculus.
These tools are not replacing all software engineers overnight, but they are dramatically increasing the productivity of individual developers, potentially reducing the total number of engineers required for a given project. The comparison to skilled trades like welding is pointed: physical, dexterous, and context-rich tasks in unstructured environments remain significant challenges for current robotics and AI, making them less immediately susceptible to full automation than certain structured digital tasks.
The Broader Conversation
Paul’s tweet taps into several concurrent threads:
- The Productivity Paradox: AI is making developers more productive, which could compress team sizes while increasing output.
- The Skill Pivot: The value may shift from writing raw code to higher-order skills: system architecture, prompt engineering for code generation, AI integration, and managing AI-assisted development workflows.
- The Physical-Digital Divide: It highlights the perceived asymmetry in AI progress, where digital domains (text, code, images) are advancing faster than generalized physical robotics.
The tweet itself is a observation, not a research paper. It provides no data or new technical findings. Its power lies in crystallizing a widespread industry concern into a single, memorable contrast.
gentic.news Analysis
Rohan Paul’s viral comment is less a prediction and more a reflection of a trend already in motion, one we've been tracking closely. This aligns with our December 2025 coverage of Anthropic's Claude 3.5 Sonnet, which set a new bar for coding proficiency on benchmarks like SWE-Bench, and our analysis of Microsoft's GitHub Copilot enterprise adoption data, which showed measurable reductions in time-to-task completion for developers.
The core of Paul's argument—the contrast between digital and physical labor—merits scrutiny. While it's true that large language models have made astonishing progress on code, the narrative that physical trades are "safe" is potentially oversimplified. As we noted in our feature on Figure AI's humanoid robots and Tesla's Optimus updates, the frontier of embodied AI is accelerating. Welding, in particular, has been automated in controlled factory settings for decades. The harder challenge is the on-site, adaptable work, but the economic incentive to automate is immense. The more nuanced takeaway is that highly structured cognitive work (like writing boilerplate code or translating specs to functions) is being automated first, while tasks requiring complex physical-world sensing and manipulation or truly novel, creative problem-solving remain harder.
This tweet also connects to the strategic pivot we're seeing in major tech education platforms. Coursera and Udacity have been rapidly adding "AI for Software Engineering" and "LLM Integration" courses over the past 18 months, implicitly acknowledging that the pure coding curriculum is no longer sufficient. The advice isn't to stop learning to code, but to learn to code in the age of AI—a fundamentally different skill set emphasizing oversight, evaluation, and integration.
Frequently Asked Questions
Is AI really going to replace software engineers?
In the short to medium term, AI is more likely to augment software engineers than replace them entirely. However, it is increasing individual productivity, which could lead to smaller engineering teams for the same output. The role is evolving from writing every line of code to designing systems, crafting precise instructions for AI coders, and reviewing and integrating AI-generated code. Entry-level programming tasks are the most susceptible to automation.
Why are trades like welding considered less vulnerable to AI?
Paul's point hinges on the current state of AI. While software operates in a pure digital realm, easily interpreted by AI, welding is a physical task requiring dexterity, real-time adaptation to materials, and operation in unpredictable environments. While robotic welders exist in factories, the cost and complexity of a general-purpose robot that can work on a construction site like a human are still prohibitive. AI's progress in software has simply been faster than in generalized robotics.
What should aspiring developers learn if AI is automating coding?
Aspiring developers should focus on skills that complement AI, not compete with it. This includes:
- System Design & Architecture: Understanding how to build large, scalable, and maintainable systems.
- Problem Decomposition & Specification: The ability to break down complex problems into clear, implementable steps for an AI.
- Prompt Engineering for Code: Mastering how to instruct LLMs to generate the correct, efficient, and secure code.
- Security & Code Review: Critically evaluating AI-generated code for bugs, vulnerabilities, and inefficiencies.
- Domain Expertise: Combining coding knowledge with deep understanding of a specific field (e.g., biology, finance).
Was the "learn to code" advice wrong?
Not necessarily, but it's incomplete. Coding literacy remains a powerful and valuable skill. The error was in presenting it as an automatic ticket to lifelong career security. The new advice is to learn to code while also understanding the AI tools that are reshaping how code is created, making the developer a manager and architect of AI productivity.
