To those who fired or didn't hire tech writers because of AI

Original Article ↗ Hacker News Discussion ↗

Scope of AI’s Impact on Tech Writers and Engineers

  • Several commenters argue you now need fewer writers or engineers per company, but not zero; AI lets 1 person plus tools replace part of a former team.
  • Others counter that this logic eventually applies to most knowledge workers, not just writers, and society is unprepared for the scale of change.
  • Disagreement over macro outcomes: some expect total employment to stay similar but spread across more, smaller companies; others think capital will simply cut labor overall.
  • A side debate explores “zero labor cost” scenarios: one camp says that would explode demand for work; another notes real organizations don’t behave like pure economic models.

What Good Technical Writers Actually Do

  • Many describe strong tech writers as anthropologists or usability radars: they bridge engineers, PMs, support, and users and often improve the product itself.
  • They act as stand‑in users, running procedures end‑to‑end, surfacing unclear workflows and mismatched mental models.
  • Key skills cited: deciding what to document, detecting assumed knowledge, prioritizing user pain, and separating major usability issues from minor ones.
  • Good writers often gather new information rather than just reformatting existing text: interviewing experts, probing edge cases, testing real systems, and building trust with audiences.

Limits and Failure Modes of LLM‑Generated Documentation

  • Hallucinations remain common and subtle: fabricated APIs, invented methods, or incorrect flows that compile or “look” right but fail at runtime or in practice.
  • Critiques emphasize lack of judgment: models don’t know what’s important, what’s unstable, what needs warnings, or when docs contradict reality.
  • Several note LLM prose is verbose, bland, and slightly incoherent at a higher level; once readers detect “LLM-isms”, they mentally tune out.
  • A recurring concern is long‑term “slop”: tiny hallucinations and misassumptions accrete into polluted codebases and documentation that misleads both humans and future AIs.
  • LLMs also can’t actually use products or “feel” confusion; they only remix what’s already written, so they cannot replace user‑experience–driven discovery.

Where AI Works Well (and For Whom)

  • Many report success using LLMs to:
    • Turn engineer-written context into readable drafts following a style guide.
    • Improve grammar and clarity for non‑native English writers.
    • Auto‑generate mediocre but better‑than-nothing docs for projects that previously had none.
  • Some teams already rely heavily on AI for site copy, READMEs, or internal docs, with humans shifting to editorial and verification roles instead of first-draft writing.
  • Others argue today’s AI may already outperform “average” or contract tech writers who produce expensive, low‑value word salad.
  • A minority believes upcoming “agentic” systems plus retrieval over existing docs will match or beat human documentation for most mainstream products.

Quality vs Cost, Incentives, and ‘Enshittification’

  • Multiple comments frame the shift as classic “quality extraction”: 50% quality at 10% cost is seen as rational by management, especially in low‑competition or captive markets.
  • Some note documentation has already degraded for decades as professional writers were cut; AI is just the newest justification in an existing downward trend.
  • Observations from transit and other sectors: bad docs rarely cause obvious, traceable revenue drops, so cuts look safe on paper even as user experience quietly decays.
  • Others worry about a “cartel of shitty treatment”: users as resources, self‑help everything, and future customer service mediated entirely by bots and AI docs.

Debates About Roles and Skills

  • One camp insists “technical writing is part of software engineering” and specialized writer roles, like testers or DBAs, were always destined to shrink.
  • Pushback: specialization still matters; average engineers are poor at audience analysis, structure, and empathy, and high‑quality docs are “night and day” better when written by pros.
  • Some writers emphasize that their real work is observation, empathy, and curation of truth under uncertainty; dismissing this as “just writing words” is seen as fundamentally misunderstanding the job.
  • Others concede AI will replace many mediocre writers, but argue that strong writers who learn to orchestrate AI will remain highly leveraged and in demand.

Broader Reflections on Writing, AI, and Human Skill

  • Several commenters are deliberately trying to improve their own writing despite (or because of) AI, viewing writing as critical thinking and “brain‑shaping” that tools can’t replace.
  • Many express fatigue with uniform LLM style; they now actively scan for and avoid AI‑generated text.
  • AI editing tools are seen as helpful for surface‑level grammar, but weak at deeper tasks like structure, persuasion, and emotional impact—areas where human editors still shine.