TC Sottek's piece at The Verge this week crystallizes something that's been accumulating: the better AI gets at demos, the more clearly it reveals the gap between demonstration and utility. Colleagues went hands-on with Google's new Gemini agent, Spark, and their takeaways were, to use Sottek's framing, sobering. Separately, The Atlantic ran a piece categorically rejecting AI consciousness claims, arguing the line of thinking is absurd. Both pieces are circling the same structural truth: AI's most persuasive quality is its appearance of capability, not the capability itself.
The Demo Is the Product
Google's Gemini Spark is the latest in a lineage of AI agents that impress in controlled conditions and flatten in real use. The Verge hands-ons noted the familiar pattern: smooth in narrow tasks, erratic at the edges. This is not a bug in a product. It is a feature of how AI is currently funded and marketed. A 2026 arXiv paper on visual graph scaffolds for LLM structured reasoning by Lei, Xiao, and Wei identifies a precise technical version of this problem: LLMs perform better on structured reasoning tasks when given external scaffolding, meaning their apparent reasoning capability is partly a performance of context, not a stable underlying competence. The demo environments are scaffolded. The real world is not. TurboFund's piece on not spray-and-praying investor outreach is accidentally a useful AI analogy: impressive surface behavior that hasn't been tested for depth tends to collapse under scrutiny.
Consciousness Claims and the Attention Economy
The Atlantic's philosophy piece on AI consciousness makes a simple argument: attributing inner experience to language models is a category error that reveals more about human projection than machine capability. The piece lands the week that Hatta's arXiv paper on AGI-oriented reproducibility argues we need new legal frameworks for AI that treat verifiability, not performance, as the core requirement. Together they point at the same failure mode: an industry built around demonstrated performance rather than verified capability, consciousness discourse as its cultural superstructure, and a press cycle that amplifies both. PlayStation's return to single-player narrative games after costly live-service failures is the adjacent story that offers a template: when the spectacle fails to retain users, you go back to the thing that actually works.