Microsoft's Copilot chatbot is running into problems

Original Article ↗ Hacker News Discussion ↗

AI-first vs User-first, and Enterprise Incentives

  • Many see “AI-first” as “shareholder-first,” not “user-first.”
  • Commenters stress Microsoft’s real customers are CIOs and procurement, not end users; captive enterprise bases let them push AI regardless of usability.
  • Some describe internal bans on non‑Microsoft AI tools and procurement rules that forbid even evaluating competitors, suggesting numbers and lock‑in matter more than user value.

Adoption, Churn, and Competitive Position

  • Cited figures: ~3.3% of M365 users pay for Copilot, and primary usage among subscribers is reportedly falling while Gemini rises.
  • In side‑by‑side trials, workers often choose ChatGPT or Gemini instead. Some enterprises only actively use a small fraction of paid seats, implying significant churn and buyer’s remorse.
  • Several note Microsoft marketing claims of unprecedented “growth” without concrete numbers, comparing it to past hype cycles.

Product Quality, Fragmentation, and UX Failures

  • A recurring theme: Copilot isn’t limited by model quality but by sloppy productization.
  • Users report:
    • Outlook/Teams Copilot that can’t actually act on mail or meetings, or returns only generic tips.
    • Excel/Office integrations that don’t understand the open document.
    • Azure Copilot giving wrong or useless answers, or failing at trivial tasks.
    • Hard limits and obvious bugs (e.g., only seeing first page of email results).
  • Multimodal features (image understanding, document generation) are described as “a disaster.”
  • Copilot is often called a censored, dumber skin over OpenAI models. Some inside Microsoft reportedly prefer Anthropic tools.
  • Branding is seen as chaotic: ~30 “Copilot” products, Office rebranded as “Microsoft 365 Copilot,” conflicting Copilot experiences that don’t interoperate.

Forced Integration, KPI Culture, and Backlash

  • Users resent Copilot buttons “sprinkled everywhere” in Windows, Office, Edge, and Azure without real integration, likened to adware/“free toolbars.”
  • People expect preferences (e.g., disabling AI, past Windows 10 push) to be overridden by updates, given Microsoft’s history.
  • Several attribute this to OKR/KPI pressure: success measured by “AI attached” and tokens consumed, not by accepted suggestions or real productivity gains.

Structural and Strategic Issues

  • Disorganized enterprise data silos and messy, conflicting content are seen as a fundamental blocker for meaningful AI assistants.
  • Commenters compare this to earlier Microsoft missteps: Windows Phone, Bing chat, .NET‑everywhere branding, Google+‑style bundling, and a general pattern of launching early, then mis‑marketing and internally fragmenting promising tech.
  • Some think AI will eventually be infrastructure like electricity; others frame the current Copilot push as an AI bubble where stock price, not product-market fit, is the true “product.”