Doctorow: 'loyal user agents' are the cure for enshittification

Cory Doctorow's Pluralistic essay is ostensibly about web browsers, but it reads like a pre-mortem for the AI assistant era. His argument is that the internet got worse not because of any single villain, but because the constraints that used to discipline tech firms (competition, regulation, rival tools like ad-blockers and mods, and workers willing to refuse) all weakened at once. When the fear is gone, what he calls 'mediocre sociopaths' run the product roadmap.

The organizing idea he wants back is the 'user agent.' In the original web standards sense, your browser is supposed to be your agent: it fetches the page on your behalf and renders it according to your preferences, including blocking things you do not want. He contrasts 'faithful' agents (a browser that warns you off a malicious site) with 'faithless' ones (smart speakers that record, apps that track location without consent, devices that phone home every few minutes). The W3C's own Priority of Constituencies says 'user needs come before the needs of web page authors,' and Doctorow argues the 2017 EME standard was the moment that principle was openly broken, because it let streaming services block accessibility tweaks that colorblind or seizure-prone users actually needed.

The sharpest part of the piece is about apps versus the open web. Roughly half of web users run an ad-blocker, which he calls the largest consumer boycott in human history, and that single fact is why web business models still have to negotiate with users at all. Apps escape that discipline because modifying them runs into DMCA Section 1201, with penalties he cites as up to five years in prison and a $500,000 fine. He uses Louis Barclay's 'Unfollow Everything' tool, which Facebook threatened off the internet, as the canonical example of a faithful agent killed by law rather than by code.

The constructive turn leans on Robin Berjon's paper 'The Fiduciary Duties of User Agents,' which proposes making the loyalty obligation legal rather than aspirational, in the spirit of attorney-client or doctor-patient duties. He pairs that with EU Digital Markets Act-style interoperability, so users can leave (his example is moving between WhatsApp and Signal without losing contacts), and with Seth Schoen's older 'owner override' idea, on the logic that letting users override their devices has historically gone worse less often than letting vendors override their users.

The honest caveat is that the essay is an argument, not a policy paper: it does not work through how a fiduciary duty would be enforced against a probabilistic system, or how 'owner override' should resolve on a shared device, and it predates the current wave of AI agents booking flights and reading inboxes on our behalf. But the frame travels. The next assistant you let act on your behalf is a user agent in exactly Doctorow's sense, and the right first question to ask it is whose side it is on.