AI companies are pivoting from creating gods to building products
AI’s shift from lofty promises of “god-like” general intelligence to more grounded, product-focused applications is prompting scrutiny of where generative models actually add reliable value. Commenters highlight the core technical tension: traditional software is deterministic and safely composable, while LLMs are probabilistic, error-prone, and hard to verify, making them best suited for assistive roles, creativity, and fuzzier tasks rather than critical automation. Many see current “AI-first” branding as reminiscent of past hype cycles (.com, blockchain), arguing that durable businesses will treat AI as just one tool in solving real user problems rather than the product in itself.
Determinism, Reliability, and Where AI Fits in the Stack
- Many argue generative models are too nondeterministic to serve as foundational components: you can’t reliably “stack” systems on top of outputs that are wrong 5–10% of the time.
- Others respond that traditional software is not perfect either, but critics counter that ordinary code failures are orders of magnitude rarer and more predictable.
- Suggested pattern: use AI for suggestions, summarization, and statistical tasks (e.g., sentiment over thousands of reviews), then hand off to deterministic systems for critical actions (payments, bookings).
- Some see opportunity in domains where verification is automatic (e.g., test generation that only counts compilable, runnable tests that improve coverage).
Productization vs. “We’re an AI Company” Hype
- Strong sentiment that companies are starting from “we have AI, now find a product,” similar to earlier “.com,” “mobile,” and “blockchain” bubbles.
- Several commenters argue AI should be treated like any other tool (like Python), not the core identity of a company.
- Others defend tech-first exploration: for large technological shifts, it can be rational to ask “what business can ride this wave?” even before specific user demand is clear.
- Consensus that the market will eventually separate substantial products from shallow “AI-washed” offerings.
Chatbots and User Experience
- Many dislike generic AI chatbots embedded into websites (e.g., car dealerships), seeing them as cost-cutting measures that worsen service, similar to forced self-checkout.
- Distinction drawn between:
- Standalone assistants (like general-purpose LLM chat) that heavy users find highly valuable.
- Context-specific chatbots on sites, which often feel clumsy, misaligned with user goals, and mistrusted.
Concrete Uses and Limitations of LLMs
- Praised uses: coding help, shell/SQL snippets, parameter lookups, rough calculations, translations, brainstorming, and as a more focused alternative to web search.
- Heavy users claim dramatic time savings; they’re comfortable spotting and correcting errors.
- Others emphasize frequent hallucinations and misleading confidence, especially in factual or medical contexts, and warn against trusting outputs without verification or expertise.
- Debate over “using it wrong”: some say you must learn how to prompt and verify; critics see that as evidence AI products are still immature for general users.
AI as Augmentation Inside Products
- Some builders describe starting with a fully autonomous “AI agent” vision, then pivoting to more traditional apps where AI automates sub-tasks and humans review or control key steps.
- Common emerging pattern: “normal product with AI under the hood,” rather than “AI replaces the entire interface or workflow.”