Should AI Engineers Still Read Code in 2026? - Alex Volkov

AI Engineer21m 35s
0 comments · 0 votesOpen discussionClose discussion
Sign in to join the discussion

    Video summary

    Alex Volkov contrasts Ryan Lopopolo's view of code as an abundant intermediate product with Mario Zechner's insistence on reading critical code. The Z/L Continuum is not a permanent label for engineers: the same person may delegate a low-risk task freely while inspecting a sensitive change line by line.

    Alex Volkov uses reported engineering-output and incident data to separate speed from stability. His routing table calls for direct review of authentication, permissions, money movement and irreversible data changes; smaller reviewable pull requests; traces, evaluations and shadow mode; and separation between the builder and the verifier. Repeated mistakes should become durable guardrails rather than recurring demands on human attention.

    Alex Volkov argues that stronger models shift the review layer without removing the need for proof. Goal-seeking loops can discover tasks, execute plans and grade results, but a builder that only grades itself can hide missing review. The enduring engineering responsibility is to place judgment at the appropriate point in the system.

    Original YouTube thumbnailWatch on YouTube

    Share this page

    Alex Volkov in blue, Ryan Lopopolo in white and Mario Zechner in pale yellow beside the blue-and-white headline CODE NEEDS PROOF on black. Framed in blue with WWW.ARTIFICIAL-INTELLIGENCE.VIDEO, 10 July 2026 and duration 21m 35s.

    Alex Volkov argues that AI-generated changes need task-specific proof, reconciling Ryan Lopopolo's harness-first approach with Mario Zechner's critical-code review.