Anthropic's public calls for slowing down AI development are not altruism but a strategic move. This narrative is deployed when their products are being outcompeted by rivals like Google's Astra, serving as a pretext to pursue regulatory capture before the market realizes their competitive disadvantage.
The core business of creating AI foundation models is weak due to a lack of compounding advantage and defensibility. With leads being fleeting and tools like Open Router allowing customers to switch models instantly, companies are forced into a brutal price competition, making regulatory capture their only viable long-term strategy.
The most viable long-term business model for labs like OpenAI isn't selling commoditized tokens. It's operating like a defense company, developing secret, unreleased 'zero-day' AIs to neutralize rogue AI threats for governments and banks at a massive premium, creating a market for digital national security.
Today's AI labs use a sophisticated policy strategy: they 'embrace' the public narrative that AI is dangerous and powerful, then 'extend' it to their advantage. This justifies massive fundraising, delays IPOs, and builds the case for regulatory capture that benefits them as incumbents.
The belief that users want a single, all-in-one AI agent is flawed. A 'polyagentamorous' future is more likely, where users fluidly switch between multiple agents (like Grok, Claude, Muse) for different tasks based on their specific strengths and alignments, much like they use different apps today.
The current AI market is dominated by 'traders' seeking quick liquidity, not 'investors' building long-term businesses. This is exemplified by companies like Instinct seeking a $10B valuation weeks after launch. The focus is on flipping stock in a hot market, disconnected from fundamental business value.
The critical question for AI agents is not just safety, but 'faithful alignment.' Users will ultimately choose agents based on whether the AI is aligned with the user's personal goals or with the model company's embedded values, as seen in the functional differences between models like Claude and Grok.
A practical path to AI safety involves competing labs red-teaming each other's models before public release. This practice, standard in the cybersecurity community where firms share vulnerability data, would allow for robust, adversarial testing by the most capable teams, creating a more secure ecosystem.
The current tech era, particularly in AI, is best understood as an entertainment cycle. From the drama of competing labs and doomsday narratives to the gamification of health apps, the underlying algorithm driving the industry is one that optimizes for entertainment above all else.