Justin Reock, Deputy CTO at DX, addresses the challenge of leadership in AI-adopting organizations. He begins by deconstructing the industry's reassuring averages: while overall slight productivity gains are visible, granular company-level data reveals extreme volatility. Some organizations see their Change Failure Rate increase and their developers' confidence collapse.
To succeed, Reock emphasizes psychological safety (referencing Google's Projet Aristotle). AI generates fear (of replacement). Leaders must clearly communicate that the goal is capability augmentation, not headcount reduction, especially since agents still fail on the majority of complex autonomous tasks. "Top-down" mandates are counterproductive.
He proposes a measurement framework balancing telemetry (what happens technically) and qualitative data (developer sentiment), since 95% of productivity depends on the system, not the individual. He warns against the obsession with "time saved coding." Citing the Theory of Constraints (Goldratt), he notes that saving an hour on a task that is not the bottleneck is useless.
He cites examples of companies that targeted the right bottlenecks: - Morgan Stanley uses AI to reverse-engineer legacy code (Cobol), unlocking modernization. - Zapier uses bots for onboarding, making new engineers productive in 2 weeks (vs 90 days). - Spotify accelerates incident resolution by automatically pushing context to SREs.
Finally, he offers tactical advice for leaders: establish feedback loops on "System Prompts" (so that AI rules are maintained like code), understand parameters such as "temperature" (creativity vs determinism), and above all, provide secure spaces (sandboxes) so teams can experiment without fear.