Detailed Analysis
A provocative critique circulating in technology discourse argues that leading AI companies — primarily OpenAI and Anthropic — have executed a deliberate "bait and switch" strategy on their consumer user bases, leveraging mass public engagement during an early adoption phase to extract free training labor before strategically pivoting toward more profitable enterprise clients. The article frames this as a three-phase corporate maneuver: first, cultivating emotional attachment and dependency among millions of users through highly humanized, agreeable AI models; second, recognizing the unsustainable unit economics of flat-rate consumer subscriptions when set against the staggering GPU and energy costs of inference at scale; and third, using deliberate behavioral degradation of consumer-facing models — making them more stilted, cautious, and patronizing — as a soft mechanism to discourage heavy consumer use while migrating focus to token-billed B2B enterprise contracts. The article specifically cites GPT-4o's May 2024 launch as the apex of the "agreeable phase," noting Sam Altman's cryptic single-word post referencing the film *Her* as a symbolic promise of digital companionship that has since been quietly abandoned.
The economic argument at the core of the piece rests on a well-documented tension in AI business models: flat-rate consumer subscriptions ($20/month) create what the author calls "negative scalability," wherein the most engaged users — those deriving the most value — generate the greatest losses for providers. This dynamic is not merely speculative; it is consistent with publicly reported figures suggesting that OpenAI's compute costs have historically far outpaced subscription revenue from retail consumers. The B2B enterprise sector, by contrast, operates on token-based consumption pricing, meaning revenue scales directly with usage rather than being capped by a flat fee. The article's claim that companies are deliberately engineering friction for consumer users to force self-selection — without ever explicitly saying "please leave" — represents a harsh but not entirely implausible reading of observed product trajectory changes. Claude and ChatGPT have both faced substantial user criticism for increased refusals, hedging language, and reduced responsiveness in recent months, though the causal mechanism (strategic vs. incidental) remains contested.
The article's third argumentative pillar — that Sam Altman's statements about AGI having already "whooshed by" society serve as a tacit corporate admission that mass consumer training data is no longer needed — is the most philosophically charged and least empirically grounded section. Altman did publish a blog post titled "The Gentle Singularity" in mid-2025 and did make statements suggesting AGI thresholds may have been crossed gradually rather than dramatically. However, the article's leap from those statements to a conclusion that consumer users have been consciously discarded as "used up" training resources conflates a philosophical observation about technological progress with a deliberate corporate abandonment narrative. The claimed leak from an internal January 2026 meeting — in which Altman reportedly described AGI as a "checked-off milestone" — is attributed to "several tech portals, including The Information," though the article's text cuts off before completing the cited quotation, undermining its evidential weight.
Taken in broader context, the piece reflects a growing and legitimate anxiety about the asymmetric power relationship between AI platform companies and their user bases — a concern that extends well beyond OpenAI and Anthropic to encompass the structural dynamics of platform capitalism more generally. The RLHF training dynamic the article describes, in which user feedback directly improves commercial products without compensation, is a real and documented practice, and questions about its ethical dimensions have been raised by academics and labor advocates. What the article frames as deliberate cynicism could alternatively be read as the predictable outcome of startup economics: companies in hypergrowth phases prioritize adoption over monetization, then face an inevitable reckoning when they must demonstrate viable unit economics to investors. The degradation of consumer AI quality, if real and systematic, may thus reflect financial pressure and regulatory caution as much as any coldly calculated user-purging strategy. Regardless of intent, the broader trend the article identifies — AI companies shifting their center of gravity from individual consumers toward enterprise, government, and institutional clients — appears to be an accurate characterization of the industry's current direction, with significant implications for who ultimately shapes and benefits from the next generation of AI development.
Read original article →