Skip to content

AI Enthusiasts Are in a Race Against Time, AI Skeptics Are in a Race Against Entropy

Source: AI Enthusiasts Are in a Race Against Time, AI Skeptics Are in a Race Against Entropy by Charity Majors

TL;DR

Charity Majors argues that the widening chasm between AI enthusiasts and skeptics in engineering teams is rooted in both sides being right — enthusiasts see real discontinuous capability leaps, while skeptics see reliability degrading and institutional knowledge evaporating. The Fin/Intercom case study (3× PRs/headcount in 9 months) serves as the North Star, achieved through exceptional engineering discipline rather than AI magic. Her framework: fix shared reality by telling the whole story, treat it as an engineering problem not a rhetorical one, double down on discipline (AI amplifies existing culture), and earn credibility by owning consequences. Key results: 3× output, defect backlog shrunk >50%, downtime down 35%.

The Chasm

The debate between AI enthusiasts and skeptics in engineering is not a disagreement over facts — it is a disagreement over which facts matter. Enthusiasts point to jaw-dropping capability leaps: models that write production code, debug complex systems, and automate tasks that required senior engineers six months ago. Skeptics counter with reliability degradation, context collapse, and the slow evaporation of institutional knowledge as junior engineers offload thinking to LLMs.

Majors's central insight: both sides are right. The enthusiast sees the possible future; the skeptic sees the fragile present. The problem is not that either camp is wrong — it's that they are talking past each other, and the organisation suffers for it.

The Fin/Intercom North Star

The most compelling data point in the piece comes from Fin, Intercom's AI customer support agent. Over 9 months:

  • 3× PRs per headcount
  • Defect backlog shrank >50%
  • Downtime fell 35%

These are not startup numbers — these are enterprise-scale metrics from a company with serious SRE discipline. Majors is emphatic that this outcome was not achieved through AI magic but through exceptional engineering discipline applied to AI adoption. Fin's team treated AI as an integration problem, not a replacement strategy. They measured everything, maintained high code quality standards, and kept humans-in-the-loop for validation.

The lesson: AI amplifies existing culture. If your engineering culture is disciplined, AI makes you faster. If it's chaotic, AI accelerates the chaos.

The Framework: Four Steps to Shared Reality

Majors proposes a concrete framework for bridging the enthusiast-skeptic divide:

1. Fix Shared Reality — Tell the Whole Story

Enthusiasts share success stories; skeptics share horror stories. Neither is lying, but neither is telling the full picture. Leaders must create a shared fact base that includes both the wins and the failures.

2. Treat It as an Engineering Problem, Not a Rhetorical One

The debate should not be about whether AI is good or bad, but about specific engineering trade-offs. What are the latency budgets? What are the correctness requirements? What is the blast radius of a mistake?

3. Double Down on Discipline

AI tools surface every weakness in your engineering practices. If your code review is inconsistent, AI-generated code will be inconsistent at scale. If your tests are flaky, AI-generated test suites will be flaky squared. The only winning move is to tighten discipline — AI forces you to become a better engineer or reveals you as a worse one.

4. Earn Credibility by Owning Consequences

Both enthusiasts and skeptics should be measured by outcomes, not predictions. Did the AI integration actually reduce pager load? Did it actually ship features faster? Measure, publish, iterate.

Key Takeaways

  1. The chasm is real — enthusiasts see discontinuous leaps, skeptics see decaying reliability. Both perspectives are valid.
  2. Discipline precedes magic — the Fin/Intercom results came from exceptional engineering practice, not AI wizardry.
  3. AI amplifies culture — good engineering cultures get better; bad ones get worse faster.
  4. Fix shared reality — stop arguing narratives and start sharing data. Build a common fact base.
  5. Own the consequences — measure outcomes, not intentions. Let results settle the debate.