AI Engineers: Should You Still Read Code in 2026?

In the rapidly evolving world of artificial intelligence, a critical question looms for engineers: Will the ability to read and write code remain a core skill in 2026 and beyond? Alex Volkov, host of ThursdAI, tackled this provocative question at the AI Engineer World's Fair, presenting insights that challenge conventional wisdom about the future of AI development. The discussion, framed around the 'Z/L Continuum,' explores how the increasing capabilities of AI agents are reshaping the role of human engineers.

The Shifting Landscape of AI Engineering

Volkov introduced the concept of the 'Z/L Continuum,' a spectrum that categorizes the engagement level of AI engineers with the code generated by AI agents. On one end, labeled 'Z,' are engineers who meticulously read and inspect every line of code produced by AI. On the other end, labeled 'L,' are those who trust the AI's output implicitly, with minimal human oversight. The presentation highlighted a significant shift within the industry, suggesting that AI engineering is moving towards the 'L' end of this continuum.

Data presented from sources like GitHub indicated a dramatic acceleration in development cycles. For instance, the number of pull requests merged and new repositories created per month has surged, with a significant portion of this activity attributed to AI assistance. This trend suggests that AI is not just a tool but a fundamental partner in the software development process, capable of generating code at an unprecedented scale and speed.

The Rise of the 'Agent Babysitter'

A key theme emerging from the discussion was the evolving role of the AI engineer from a direct coder to more of a supervisor or 'agent babysitter.' Volkov cited statistics indicating that for some, like Boris Cherry, a creator of Claude at Anthropic, nearly 80% of their code is now AI-written, with minimal human edits. This shift implies that the engineer's primary responsibility is moving towards ensuring the quality, safety, and alignment of AI-generated outputs rather than the manual construction of code.

The implications of this change are profound. If AI agents can efficiently handle the bulk of code generation, the critical human skills may lie in prompt engineering, system design, and the ability to guide and validate the AI's work. This transition raises questions about the future demand for traditional coding skills and the new competencies that will be valued in the AI engineering workforce.

The 'Z/L Continuum' and the Future of Proof

Volkov presented a framework that helps categorize where engineers currently stand on this continuum, posing the question: 'Where are you?' The continuum ranges from 'Z' (Zecher, 'slow the f*** down,' implying deep code inspection) to 'L' (Lapopolo, 'code is a liability,' suggesting minimal human intervention). He noted that while many engineers might still be closer to the 'Z' end, the industry trend is clearly pushing towards 'L' as AI capabilities advance.

The discussion also touched upon the idea that 'the continuum is not people, it is tasks.' This perspective suggests that different tasks might require different levels of human oversight and validation. For instance, critical tasks involving sensitive data or high-stakes decisions might still necessitate rigorous human review, while less critical tasks could be delegated more readily to AI agents. The key, Volkov implied, is to understand which tasks require human judgment and which can be automated.

The Emergence of 'Loops' in AI Engineering

A significant aspect of the future landscape highlighted was the concept of 'loops' in AI systems. Peter Steinberger, creator of Spectacle, and Boris Cherry from Anthropic both emphasized the shift from prompting coding agents to designing 'loops' that prompt their own agents and then verify the outputs. This suggests a move towards more autonomous and self-correcting AI systems, where human engineers act as architects of these intelligent systems rather than direct executors.

The idea of 'loops' implies a continuous cycle of task discovery, planning, execution, and verification, with AI agents playing a central role. If the AI builder grades its own work, the human role shifts to ensuring the integrity of that self-grading process and the overall system design. This represents a significant evolution in how software is built and maintained in the age of advanced AI.

The Critical Question: Proof and Certainty

The presentation concluded by posing a critical question for the future: 'What proof does this change need?' Volkov highlighted that while AI can accelerate output, the challenge lies in ensuring the quality and reliability of that output. He cited data showing a significant increase in code deletion ratios and incidents per pull request when human review is bypassed, suggesting that while AI can speed up development, it also introduces new risks if not managed properly.

Ultimately, the discussion underscored that the future of AI engineering is not about eliminating human oversight but about redefining it. As AI capabilities continue to advance at an unprecedented pace, engineers will need to adapt, focusing on higher-level tasks like system design, prompt engineering, and ensuring the responsible and effective use of AI tools. The key takeaway is that while AI may handle the 'how,' humans will remain crucial for the 'what' and 'why' of software development.