Agentic Web Realizes Berners-Lee's Semantic Vision — But Centralizes Power in Corporate AI Labs, Displacing Human Agency and Indigenous Data Sovereignty
Original framing: “The agentic web is here” — MIT Technology Review
Kenyan data annotators for Sama (OpenAI's contractor) earn $2/hour filtering agentic training data while suffering PTSD from toxic content review — their labor makes 'autonomous' agents possible. Brazilian gig workers on iFood platform face algorithmic management by agentic dispatch systems they cannot audit. Disability advocates warn that agentic interfaces assuming neurotypical communication patterns exclude autistic users who navigate differently. The original article cites zero affected communities; its 'we' is Silicon Valley engineers and early adopters.
The agentic web's arrival is not the fulfillment of Berners-Lee's vision but its inversion: where he designed decentralized standards (RDF, SPARQL, OWL) governed by W3C's multi-stakeholder process, we now have centralized foundation models controlled by four corporations whose business models require behavioral surplus extraction.
This shift was not technically inevitable — it was driven by compute economics that favor monopoly, a history the MIT Technology Review narrative erases by presenting agentic systems as natural evolution. Restoring the original vision requires three simultaneous interventions: protocol-level interoperability mandates (Solid), Indigenous data sovereignty as legal precedent, and public compute infrastructure to break the hardware monopoly. Without all three, the agentic web becomes a new enclosure movement — fencing the open web's knowledge commons into proprietary agentic services that mediate human intention through opaque, unaccountable, and ecologically costly intermediaries. The trickster sees the joke: we were promised agents that serve us; we got agents that serve the firms that own them.