Cloudflare's AI Psychosis

Original Article ↗ Hacker News Discussion ↗

Cloudflare’s recent pivot toward rapid AI- and developer-feature releases is drawing mixed reactions, with critics arguing that product sprawl, weaker documentation, and more frequent outages signal a loss of focus on core infrastructure reliability. Others counter that at hyperscale some failures are inevitable and that aggressive experimentation is necessary to stay competitive in an AI-driven market. The exchange broadens into a critique of how big tech balances speed vs. robustness, the cultural role of product management vs. engineering, and the long‑term risks of centralizing so much of the web’s traffic behind a few large platforms.

Reliability, Outages, and Expectations

  • One line of argument: infra providers should have at most 1–3 outages per year; more suggests poor engineering/leadership and justifies moving off to self‑hosting.
  • Counterpoint: outages are an inevitable class of bugs, especially at scale; demanding near-zero incidents is unrealistic even for AWS or aviation‑style safety‑critical systems.
  • Debate over analogies: some compare software outages to collapsing bridges; others argue real‑world transport systems also fail regularly yet remain essential.
  • Some users report serious impact from CF and GitHub outages (deploys blocked, OTA updates failing), prompting reconsideration of reliance on large infra vendors.

AI, Velocity, and “Psychosis”

  • Several see CF’s AI push as part of a broader industry pattern: everyone is or will be AI‑augmenting workflows, and blaming AI alone for bad products is misplaced.
  • Pro‑AI voices emphasize “obscene” productivity gains: huge code changes, rapid docs/test generation, and self‑healing systems.
  • Critics argue this often produces low‑quality “slop,” overwhelms maintenance capacity, and burns planetary and social resources for questionable value.

Product Sprawl, UX, and Strategy

  • Many agree CF launches too many overlapping services (D1, Durable Objects, KV, R2, Queues, Hyperdrive), leading to confusion and perceived lack of coherence, even if each has a logical niche.
  • Complaints about poor docs, wrong client libraries, and tools like wrangler being powerful but unwieldy.
  • Some defend CF as one of the few genuinely innovating infra players; Workers, Durable Objects, and free‑egress R2 are singled out as strong offerings.
  • Another criticism: features ship fast, then get neglected; “abandonware” concerns mirror those for AWS/GCP.

PM vs. Engineering Culture and Market Pressures

  • Disagreement over whether “PM‑driven” culture or “engineers in charge” would produce better outcomes; some argue only customer‑obsessed leadership matters.
  • Others blame weak leadership and lax standards, not market forces, for incoherent product strategy.
  • Opposing view: in a hyper‑competitive AI market, slow, careful releases risk losing customers to faster imitators.

Centralization, Control, and the Open Internet

  • Some see CF as a “middleman cancer” threatening a free, open web via TLS termination and anti‑bot challenges; others note site owners ultimately set policies.
  • Broader concern: CDNs replacing traditional transit, raising gatekeeping risks, even while solving real DDoS and scaling problems.