AI enthusiasts are in a race against time, AI skeptics are in a race against entropy

Simon Willison recently highlighted an insight from Charity Majors that uses two elegant metaphors to dissect the internal conflict within software teams in the age of AI: AI enthusiasts are in a race against time, while AI skeptics are in a race against entropy.

Two Real Existential Threats

The enthusiasts' logic is straightforward. We've already seen teams that lean heavily into AI-assisted development achieve discontinuous leaps in capability—jumps that are not incremental but transformative. If you wait for the technology to stabilize, for tools to mature, for best practices to emerge, a competitor might already have captured the market with an AI-generated solution. This isn't hype; it's a genuine existential threat—the window of opportunity may close before the dust settles.

The skeptics' fear is equally concrete. When code is shipped faster than engineers can read it, and no single person fully understands the context of a given module, you are making withdrawals from a trust account that took years to build. System reliability degrades, institutional knowledge evaporates, and on-call rotations become a waking nightmare. Eventually you are left with an incomprehensible product and a burned-out team. That too is a real existential threat.

Speed vs. Quality Is Not a Zero-Sum Game

On the surface, this looks like the classic speed-versus-quality tension. But Charity’s key insight is that both perspectives are "not wrong." This means the problem isn't about choosing sides; it's about the broken communication between two valid worlds.

Within a team, the benefits that enthusiasts experience are real and tangible: a feature goes from description to working prototype in minutes. The costs that skeptics see are equally real: AI-generated code arrives with subtle bugs, poor error handling, and messy abstractions—yet it's already merged. Each side operates within its own reality, making decisions based on that reality, but there is no bidirectional information channel linking them.

The Missing Feedback Loop

This brings us to Charity’s core diagnosis: “There is no natural feedback loop connecting enthusiasts with skeptics.” In traditional software engineering, mechanisms like code review, testing, and production monitoring naturally form a feedback system. But those mechanisms are tuned to the pace of human development. When AI increases the speed of generation by orders of magnitude, the existing feedback loops are overwhelmed.

This is not a technical problem but an organizational design challenge. You need to deliberately construct new feedback loops that allow “lessons from speed” to flow into “quality assurance processes,” and let “findings about stability” feed back into “how we use AI.” For example, you could regularly feed bug patterns from AI-generated code back into prompt engineering; or hold dedicated “AI output retrospectives” where enthusiasts and skeptics sit together and discuss based on real data, not impressions.

Trends and Implications

This reveals a broader shift: software engineering is moving from a “manufacturing” paradigm to a “management” paradigm. In the past, engineers spent most of their time writing code. Now an increasing share of work involves managing AI inputs, verifying AI outputs, and maintaining overall system comprehensibility. This shift demands an organizational upgrade—a fusion of engineering and management capabilities at a higher level.

For tech leaders and individual engineers, the takeaway is clear: avoid the trap of choosing a side. The truly valuable work is actively building those missing feedback loops. It could be a lightweight dashboard tracking defect rates and rollback rates for AI-contributed code. It could be a regular exchange where both sides share real case studies. Whatever the format, the goal is to close the widening gap in “shared reality” between the two groups.

In the AI era, competitive advantage may not just come from how fast you deliver features. It may hinge on whether you can build these internal feedback loops faster than your rivals—embracing speed while taming entropy, so that both racers ultimately move in the same direction.