Not everyone is using AI for everything

Original Article ↗ Hacker News Discussion ↗

Overuse and Misfit of LLMs

  • Many report companies replacing simple, deterministic workflows (support flows, CI, tooling, code review) with slower, less reliable LLM-based systems, often just to say they “do AI.”
  • Some contractors say they regularly implement unnecessary AI because executives insist, even when PoCs show deterministic designs would be cheaper, faster, and more stable.
  • A recurring critique: AI is being used to replace tools it should instead be used to build (e.g., generating lint rules or scripts rather than acting as the lint/CI itself).

Deterministic Systems vs Agents

  • Strong camp arguing core business logic and customer-facing workflows must be deterministic and testable; LLMs should be wrappers or assistants around robust CLIs/APIs.
  • Others promote hybrid patterns: agents reason in natural language but must execute via carefully designed, constrained tools.

LLMs for Coding

  • Many developers use LLMs heavily as coding assistants: boilerplate, one-off scripts, refactors, explanations, debugging.
  • Supporters claim huge speedups; critics say LLM-generated code is brittle, insecure, overengineered, and harder to understand, often costing more time in review and debugging.
  • Several warn of eroding fundamental skills and future codebases that are effectively “software archaeology” projects.

Workplace Pressure and Management FOMO

  • Reports of AI hackathons, daily-use mandates, and bonus structures tied to token usage, even where a 1–2% failure rate is unacceptable.
  • Engineers often see management chasing investor narratives and “AI adoption” metrics rather than measurable product value.

Who Is Actually Using AI?

  • Thread cites studies: ~20–30% of US working-age population using AI tools regularly; far below “everyone,” but high for a new tech.
  • Debate over definitions: explicit chatbot use vs. passive use embedded in Google search, recommendation feeds, phone cameras, etc.
  • Some argue “everyone uses AI” if you count those background systems; others see that as forced or incidental, not meaningful adoption.

Individual Use Cases and Non-Use

  • Heavy users describe substantial real-world benefits (insurance disputes, home repair decisions, garden design, shopping, marketing).
  • Others restrict use to programming or search replacement; some avoid LLMs entirely and feel pressured or marginalized.

Hiring and Career Anxiety

  • Job seekers see “How do you use LLMs?” as a standard interview question and struggle to answer for both AI-enthusiastic and AI-skeptical employers.
  • Some hiring managers now treat lack of agent/LLM experience as a red flag; others value nuanced, critical use over hype.

Quality, Safety, and Societal Concerns

  • Recurrent themes: hallucinations, security holes, sloppier software, degraded customer support, and “slopification” of web content.
  • Concerns that low literacy and weak critical thinking amplify misuse, while corporate incentives favor cheaper support over better service.

Long-Term Outlook

  • Optimists liken this phase to early internet or compilers: rapid growth, eventual ubiquity, and large productivity gains.
  • Skeptics foresee permanent niches where AI is ill-suited and warn against assuming future model improvements will automatically fix current structural problems.