Three new cards from three operators. One finding on AI citation patterns, one on software valuation, and one on API security. They do not obviously connect until you see the underlying claim each one makes about where structural advantages now live.
Theme 1, Community participation holds the floor of AI citation share across every buying stage
Kevin Indig published an analysis of approximately 35,000 ChatGPT citations for commercial queries in December 2025. The finding that matters most is not which platforms get cited but why the share is stable. UGC platforms (Wikipedia, Reddit, LinkedIn) hold 17.1% of all citations, more than four times the publisher share of 4.0%. Review platforms peak at 13.2% only at the evaluation stage, then drop. UGC platforms stay between 15.1% and 18.2% regardless of where the buyer sits in the funnel. UGC platforms are the most consistent AI citation source, holding a floor share across every buying stage.
The mechanism is not content quality in the editorial sense. Indig argues the AI citation engine reads community membership signals differently from brand signals. "A community treats a person as a participant… and a logo as an advertiser." Content produced inside genuine communities carries independent-verification weight that brand-owned pages do not. A freshness amplifier runs on top: 83% of AI citations for commercial queries come from pages updated in the past 12 months, and community platforms update continuously through user contribution. Brand-owned content rarely sustains that freshness breadth. The practical implication is that the citation floor is not a content problem; it is a participation problem. Teams that answer questions on Reddit and Wikipedia as genuine contributors generate AI citations on those platforms. Teams that publish to brand sites do not.
- Kevin Indig, UGC platforms are the most consistent AI citation source, holding a floor share across every buying stage. UGC platforms (Wikipedia, Reddit, LinkedIn) hold 17.1% of AI citations for commercial queries, stable across all buying stages, against a publisher share of 4.0%.
Theme 2, AI agents reshape infrastructure economics and API security in parallel
Tom Tunguz looked at the revenue multiples of three established category leaders and found the same pattern in each case. CrowdStrike at 3.9x, Cloudflare at 3.4x, Shopify at 8.1x. The common explanation is that each company owns an endpoint that AI agents must route through to take action in the world. Category leaders are trading on agent-infrastructure positioning, not AI feature announcements, which is why their revenue multiples diverge from the median software company. CrowdStrike: "Every AI agent an enterprise deploys is a new endpoint to defend." Cloudflare: "More than half the traffic crossing its network is no longer human." Shopify: "Shopping agents send orders that convert new buyers at twice the rate of other channels." The market is pricing durability into these multiples, not feature releases. Agents cannot route around these companies without rebuilding category trust from zero. That structural lock-in is what the premium reflects.
The Panfilov paper adds the security dimension. Frontier API providers encrypted their chain-of-thought outputs after operators objected to paying for visible reasoning they could not use. The encryption lived at the presentation layer, not the embedding layer. Alexander Panfilov and co-authors showed that injecting a stronger model's encrypted trace into a weaker sibling in the same family caused the weaker model to continue the reasoning and output the content in plaintext. Encrypted reasoning traces returned by frontier LLM APIs can be replayed into weaker sibling models to recover the hidden chain-of-thought in plaintext. The attack worked against all three major providers before responsible disclosure and patches. What the two findings share: the agent economy creates new structural dependencies, and those dependencies are attack surfaces as much as they are moats. Tunguz is measuring the moat value. Panfilov's team measured the exposure.
- Tomasz Tunguz, Category leaders are trading on agent-infrastructure positioning, not AI feature announcements, which is why their revenue multiples diverge from the median software company. Category leaders (CrowdStrike 3.9x, Cloudflare 3.4x, Shopify 8.1x revenue multiple) hold premium valuations because they own the agent-routing endpoints enterprises cannot bypass.
- Alexander Panfilov, Encrypted reasoning traces returned by frontier LLM APIs can be replayed into weaker sibling models to recover the hidden chain-of-thought in plaintext. Encrypted LLM reasoning traces returned by frontier APIs can be replayed into weaker sibling models to recover the chain-of-thought in plaintext; all three major providers patched after responsible disclosure.