Why your AI rollout is creating bottlenecks you can't see

The core argument. Elizabeth Hendrickson and Joel Tossey argue that engineering leaders must adopt systems thinking to understand why well-intentioned changes backfire. Their new book, Signals and Levers, frames organizations as non-linear adaptive systems where cause and effect are rarely straightforward. Most leaders operate on work-as-imagined rather than work-as-done, which leads to misdiagnosis and misplaced interventions.

The AI amplifier effect. AI tools are not a simple productivity lever. Hendrickson observes that AI is an amplifier that exposes bottlenecks that were previously invisible. When teams produce 300% more code and 400% more bugs, the rest of the system—code review, testing, deployment—cannot absorb the output. One tech lead described a 27,000-line pull request across 104 files that was impossible to review. The constraint shifted from writing code to reviewing it, but the organization had not adjusted.

Policy vs culture levers. The authors distinguish between two categories of levers. Policy levers require positional authority: hiring criteria, org structure, job descriptions. Culture levers are available to everyone: expressing appreciation, reframing questions, feeding what you want to grow. A leader who only uses policy levers will fail to shape culture because culture cannot be dictated. The key is to nudge rather than command.

Seeing signals before pulling levers. The most important step is observing the system without immediately intervening. Hendrickson warns that leaders often commit to timelines based on imagined work rather than experienced work. Tossey adds that dashboards show only the signals leaders expect, not the ones that matter. The real insights come from being present: understanding why velocity is volatile, why requirements are misunderstood, why the architecture is fighting the product.

The trade-off with fun. Hendrickson describes a case where engineers wanted team-building activities to compensate for a painful codebase. She refused, arguing that papering over the pain with fun activities would delay fixing the underlying technical debt. The counterintuitive move was to prioritize satisfaction through meaningful work rather than temporary relief. This requires constant signal monitoring to verify the bet is working.

What teams should watch. The authors' framework offers practical tools for diagnosing organizational friction. Teams should watch for problems that hop around between teams, metrics that tell a misleading story, and the gap between stated policy and lived experience. The real value is not in the theory but in the discipline of observing before acting. Most organizations skip this step.