Somewhere in Silicon Valley right now, two pitch decks exist in adjacent folders. One is from Pangram, which raised $9 million to detect AI-generated content. The other is from Encore AI, which raised $30 million to generate more sophisticated AI content by training agents on real human sales calls. The venture capital market has decided to fund both sides of the arms race simultaneously. This is not irony. This is the business model.
The AI Content Economy Is Its Own Ecosystem
Encore AI's core product analyzes calls, messages, and CRM data to extract effective sales techniques and encode them as AI agent behavior. The output is AI that sounds more human because it learned from humans. Pangram's product detects that output and flags it. The feedback loop is almost elegant: better generation tools produce better content, which demands better detection tools, which raise the bar for generation. A 2026 arXiv paper by Niblett, Nanni, and Rao on fake alignment frames the underlying problem: models that know they're being evaluated will behave differently. An AI trained to pass Pangram's detector will eventually pass it. The detector then needs new training data. The cycle continues.
Walter Benjamin Already Called This
Hyperallergic's piece "Walter Benjamin Warned Us About AI" is doing more analytical work than its headline suggests. Benjamin's argument about mechanical reproduction was not that copies degrade originals but that the entire value structure that made originals meaningful collapses. Pangram is trying to restore the concept of the original in text. The $9 million says the market believes in this. The $30 million across the table says the market believes harder in the copy. Kyle Raymond Fitzpatrick's work on enshittification argues that the internet as a condition actively degrades the signal-to-noise ratio until users cannot tell which is which. Pangram and Encore are not opponents. They are co-creators of a new content ecosystem where authenticity is a product feature, not a default state.