We Need Smart People Who Give a Shit
- Complex, critical infrastructure (such as banking systems like ACH, air traffic control, and major tech platforms) is inherently fragile and riddled with legacy flaws, yet it functions reliably because of "smart people who give a shit."
- These individuals rely on a self-reinforcing loop of two traits:
- Intimate knowledge of how a system works beneath the surface.
- The personal drive and sense of ownership required to patch, route around, and manually handle edge cases before things collapse.
- A recent wave of widespread outages across major institutions (including Microsoft 365, Apple, United Airlines, Air Traffic Control, the NSA, and major banks) shares a root cause: a shrinking supply of people who deeply understand and care about these systems.
- Leadership and management increasingly view generative AI as a cheaper, compliant replacement for skilled engineers because non-practitioners struggle to distinguish superficial AI output from resilient engineering.
- Generative AI lacks true intelligence, intuition, and care—it cannot sense when a system is drifting off the rails or take responsibility for novel edge cases.
- Replacing or sidelining experienced engineers in favor of "babysitting" LLMs creates a hollowing-out effect:
- Laid-off engineers take institutional knowledge with them.
- Remaining workers lose deep system comprehension and ownership because prompting distances them from the struggle of building and debugging firsthand.
- The author characterizes this trend as a "hot stove moment" for capitalism, warning that system failures will continue to escalate until leadership learns firsthand that automated output cannot replace human ownership and expertise.
Hacker News Discussion
- Capacity for maintenance: Top commenters connected the essay to Eric Hoffer's observation that a society's true test is its capacity for day-to-day maintenance rather than initial construction, noting that humans naturally struggle with maintenance because preventive work lacks short-term feedback loops.
- Comprehension debt vs. AI capability: Several engineers argued that AI isn't directly causing outages on its own, but rather accelerating "comprehension debt"—lowering the cost of generating code much faster than the cost of understanding systems, undocumented assumptions, and dependencies.
- Conversely, some users noted that AI can be useful for clearing out neglected technical debt on "side quests" when guided by someone who cares.
- Centralization and historical reliability: Commenters contrasted modern cloud outages with older systems like the Bell System's electromechanical switches and early GSM networks, arguing that modern cloud architectures prioritize time-to-market, abstraction layers, and vendor centralization ("nobody gets fired for buying AWS/Microsoft") over eliminating single points of failure.
- Management incentives and workplace politics: Many agreed that leadership prioritizes short-term margins, commoditized labor, and compliance over redundancy and robustness.
- Debate emerged over whether companies punish or reward people who care: some argued corporate environments treat proactive, outspoken engineers as liabilities, while others maintained that high-performing "A-teams" who take ownership are still heavily valued during critical escalations.