At Canva, senior engineers now spend most of their time reviewing AI-generated code rather than writing it, according to Chief Technology Officer Brendan Humphreys. In an interview with Business Insider, Humphreys revealed that engineering teams draft detailed instructions for AI agents to execute coding tasks overnight. “Often, those results are really impressive,” he said. The company’s most senior engineers now describe their jobs as “largely review”—checking output and steering agents.
Humphreys emphasized that defining problems and translating vague, confusing, or conflicting requirements into production-ready specifications remains the hardest part of engineering. Effective use of AI requires “precision of articulation” in requirements and “mastery of the domain” to verify correctness. Without deep expertise, complexity can spiral out of control across Canva’s roughly 70 million lines of code. “These tools can have you in a jungle before you know it,” he warned.
This shift extends beyond Canva. Spotify CEO Gustav Söderström said the company’s most senior developers have not written a single line of code since December, instead supervising an internal AI system. A January Anthropic report found AI appears in roughly 60% of developers’ work, but engineers can fully delegate only 0–20% of tasks. Sonar’s 2026 State of Code survey found 96% of developers do not fully trust AI-generated code is functionally correct, yet only 48% always verify it before committing.
AWS CTO Werner Vogels termed the resulting problem “verification debt”—developers spending significant additional time reviewing AI output rather than creating it. Research indicates AI-generated code shows higher cyclomatic complexity and can introduce duplicated logic and inconsistent patterns, creating technical debt visible only at scale. Senior engineers are becoming increasingly valuable for spotting architectural problems and maintaining system integrity, not for coding speed. Sonar CEO Tariq Shaukat stated the burden of work has shifted from creation to verification and debugging.
