RevenueCat Bets That Agents Will Build Your Paywalls
OpenAI Starts Watermarking ChatGPT Text — But Only in the EU. Nordic & EU: The AI Act's First Real Teeth.
RevenueCat Bets That Agents Will Build Your Paywalls
RevenueCat used its App Growth Annual event to roll out an AI Paywalls Editor that generates full paywall designs from a prompt or a screenshot, runs multi-step funnels, and predicts 12-month LTV before you've spent a dollar on traffic [4]. That's a meaningful shift from "AI helps you write copy" to "AI makes the monetization decision" — and it's shipping into a market where subscription businesses live or die on paywall conversion.
More consequential long-term is the AI Toolkit: an MCP server, CLI, skills, and plugins built so agents like Claude Code, Cursor, and Gemini CLI can manage subscriptions and build paywalls with actual codebase context, not generic templates [5]. This is orchestration infrastructure, not a feature — RevenueCat is betting that agents, not developers, will be the primary users of its platform within a few product cycles.
X chatter focused less on the paywall generator and more on agent infrastructure safety — MCP cost tracking, tools like HOL Guard for securing agent hooks [4][5]. That's the tell: the market is already past "can an agent do this" and onto "how do we not get burned when it does."
OpenAI Starts Watermarking ChatGPT Text — But Only in the EU
On October 5, OpenAI began rolling out invisible statistical watermarking ("textGrain") for ChatGPT and Codex outputs in the EU, starting in Ireland, to comply with AI Act transparency rules that took effect in August [6][7]. The watermark doesn't identify the user and is detectable only by approved parties — it's off by default everywhere else, though API customers globally can opt in [8].
This is regulation doing what regulation does: creating a two-tier product. EU users get watermarked-by-default output; everyone else doesn't, unless they ask. Google and Anthropic have comparable systems, and detection reliability drops fast on short, edited, or technical text — math, recipes, anything with low linguistic entropy [6].
The real story isn't the watermark, it's the geography. OpenAI is building compliance infrastructure region by region rather than globally, which tells you EU AI Act enforcement is being taken seriously enough to engineer around, not just lawyer around.
Nordic & EU: The AI Act's First Real Teeth
OpenAI's Ireland-first watermarking rollout is the most concrete EU AI Act compliance action we've seen from a major lab to date — a default-on transparency feature, not a toggle buried in settings [6][7]. For a region that's spent two years writing rules and comparatively little time watching labs actually implement them, this is the moment the AI Act stops being theoretical.
European regulators and users on X zeroed in on the same weak point everyone in compliance circles already knew about: detection reliability collapses on short or heavily edited text, meaning the watermark mostly catches lazy, unedited output — not determined misuse [6]. Expect this to become the template other labs either match or get pressured to match across the EU.
What This Means For Your Business
Three stories, one pattern: the unit of work is shifting from "software you configure" to "agents you direct." Perplexity's extension gives an agent hands. RevenueCat's toolkit gives agents a seat at the product-decision table, not just the code-writing table. Neither is about code quality anymore — both are about judgment: what should the agent be allowed to touch, and whose decision is it making when it builds a paywall or fills a form on your behalf.
That's the real cost center shifting. Writing a paywall generator or a browser automation script is now cheap — RevenueCat's own toolkit lets other agents write that code. What's expensive, and what you should be investing in, is the layer above it: deciding what the agent is allowed to do unsupervised, what gets reviewed, and what "good" looks like before you let an LTV model make pricing calls at scale. The OpenAI watermarking story is a preview of where this all lands — when agents produce most of the output, provenance and accountability become product features, not afterthoughts.
If you're still evaluating AI tools by "does it write good code," you're asking last year's question. The question now is whether you trust the thing enough to hand it your logged-in session, your pricing strategy, or your compliance posture — and whether you've built the judgment layer to supervise it when you do.
Key takeaway: The code is increasingly free — agents write it, generate it, ship it. The competitive edge is moving entirely to who decides what the agent should do, and who's accountable when it does it wrong.
Sources
- https://chromewebstore.google.com/detail/perplexity/pninphncmkphpjgfffammoenglghcjii
- https://www.perplexity.ai/changelog
- https://arstechnica.com/ai/2026/02/perplexity-announces-computer-an-ai-agent-that-assigns-work-to-other-ai-agents/
- https://www.revenuecat.com/blog/company/everything-we-announced-at-app-growth-annual-2026
- https://www.revenuecat.com/docs/tools/ai-toolkit
- https://arstechnica.com/ai/2026/10/openai-will-watermark-chatgpt-outputs-by-default-but-only-in-the-eu/
- https://techcrunch.com/2026/10/05/openai-will-start-watermarking-chatgpts-text-in-the-eu/
- https://www.theverge.com/ai-artificial-intelligence/1004880/openai-chatgpt-text-watermarks-eu-ai-act
- https://www.revenuecat.com/blog/company/8-updates-to-help-you-build-better-paywalls-in-revenuecat
Stay ahead of AI
No spam. Unsubscribe anytime.
Want to go deeper?
Reading the news is one thing. Exploring the frontier is another. See what we're building.