← back
Should AI Engineers Still Read Code in 2026? The Z/L Continuum — Alex Volkov, ThursdAI
Takeaway
Higher code-generation throughput makes deliberate review and engineering accountability more consequential.
Summary
- Volkov frames a tension between Ryan Lopopolo’s confidence in agent-generated implementation and Mario Zechner’s insistence on reading critical code.
- Cheap code generation shifts the scarce resource toward human attention, supervision, and verification.
- The talk contrasts increased output and agent autonomy with compounding errors and lost debugging knowledge when generated code goes unread.
- The Z/L continuum invites teams to examine their review practices, especially the distinction between critical and noncritical code.
code-reviewai-codingengineering-leadership
Original description
"How much better do the models have to get before you'll stop reading the code?" Theo asked that question recently and the replies caught fire. Mitchell Hashimoto is calling it agent psychosis. ThePrimeagen's subreddit is in open revolt about people shipping code they never read. Uncle Bob says we have about a year left of looking at code at all. Alex Volkov saw this argument coming three months ago, and gave it a name. At AI Engineer Europe, OpenAI's Ryan Lopopolo opened the conference by saying "code is free." His team shipped over 1,000,000 lines with zero human review. Mario Zechner closed the same conference telling everyone to slow the f*** down and read every line. Same stage. Opposite advice. Standing ovations for both. Alex hosts ThursdAI and spends every week talking to the people building this stuff. In this talk he lays out the Z/L Continuum: what the top AI engineers in the world actually do, not what they say on stage. Including: • Anthropic's own numbers on Claude writing 80%+ of Claude's code, and what happens when it breaks • Why human review became the bottleneck nobody wants to talk about • The uptime chart that looks like a Christmas tree • Why Dexter Horthy, the "let the agent cook" guy, publicly said "I was wrong" • The one tweet that changed how Alex thinks about this. You're not Team Z or Team L. Every task gets its own spot on the continuum, and knowing where to place it is the actual skill now. If you've ever shipped code you didn't read (be honest), this talk is about you. Speaker info: https://x.com/altryne https://thursdai.news The original Z/L Continuum essay: https://thursdai.news/zl Anthropic's "When AI Builds Itself": https://www.anthropic.com/institute/r... Lucas Meijer's tweet: https://x.com/lucasmeijer/status/2044... Timestamps: 0:00 - Introduction 0:48 - The shift in AI engineering since December 2025 1:32 - The trend of AI-assisted coding and reduced manual input 3:37 - The core conflict: "Code is free" vs. "Read every line" 6:13 - Defining the Z/L Continuum 8:02 - Analyzing the "code is free" perspective (Ryan Lopopolo/OpenAI) 9:31 - Risks of rapid AI output and incident rates 11:04 - Recursive Self-Improvement (RSI) and human review as a bottleneck 12:08 - The correction: Focusing on tasks, not people 13:56 - Recommended strategy: Routing changes for appropriate verification 15:52 - Emerging capabilities: Fable and Mythos 17:16 - Capability drift and the shift toward "Loops" 18:16 - Understanding "Loops" as the next engineering primitive 20:15 - Future outlook and maintaining flexibility #AIEngineering #AICoding #CodeReview #VibeCoding #ClaudeCode #AIAgents #ThursdAI