Subscribers to AI-powered apps cancel their annual subscriptions 30% faster than subscribers to non-AI apps. That is the finding of RevenueCat’s 2026 State of Subscription Apps Report, drawing on data from more than 75,000 app developers and over one billion in-app transactions. Annual retention for AI apps sits at 21.1%, compared to 30.7% for non-AI.

The pattern is consistent: when AI is the product, users try it, pay for it, and leave faster than they do with conventional software. But there is at least one sector where AI appears to be doing the opposite. Hostinger, one of the world’s largest hosting providers, reports that after deploying its AI agent Kodee with direct VPS terminal access, VPS retention rose from 66% to 70% and new VPS orders grew roughly fourfold.

The difference may come down to what role AI plays. In AI-native apps, AI is the product itself. In hosting, the product is still the server. AI is the layer that makes the server less painful to manage.

What Kodee Actually Does on the Server

Kodee is not a chatbot that suggests documentation links. It is an AI agent built into Hostinger’s VPS dashboard and terminal, powered by the Model Context Protocol (MCP). It can read logs, run commands, troubleshoot issues, deploy applications, configure security settings, and apply fixes directly on the server. Hostinger describes approximately 200 VPS management actions available through Kodee, with no separate subscription or installation required.

The terminal integration, shipped in April 2026, is the step that separates Kodee from advisory chatbots. A user can describe a problem in plain language, and Kodee plans the steps, executes commands, reads the output, and iterates until the issue is resolved or escalated. Over a six-month period, Hostinger logged more than 914,000 VPS conversations with Kodee.

An AI agent executing commands on live servers carries inherent risk – a misconfigured firewall rule, a misapplied fix, or an unintended deletion can take a production environment offline. Hostinger’s published materials do not describe what guardrails Kodee operates under: no public documentation covers confirmation steps before destructive commands, permission scoping, rollback procedures, or audit logging. The 7% human handoff rate suggests some escalation logic exists, and Hostinger describes the system as drawing “a clear line where a human belongs.” But the safety architecture behind an AI agent with terminal access to customer servers is a question Hostinger has not yet answered publicly.

We Tested Six Website Builder AIs With an Apple Pie Recipe. Not All of Them Are Actually AI.

The Numbers Hostinger Reports – and the Caveats It Discloses

Hostinger published its Kodee performance data on June 25, 2026. The numbers are self-reported and not independently audited. To their credit, Hostinger is transparent about the limitations:

  • 91% autonomous resolution applies to conversations with a recorded outcome. Across all 914,300 VPS conversations, including those without a definitive outcome, the rate is 69%.
  • VPS retention rose from 66% to 70% after MCP deployment.
  • New VPS orders grew roughly fourfold, but Hostinger writes: “harder to pin on Kodee alone because pricing, marketing, and other launches all happened in the same window.”
  • 7% of conversations require human handoff, rising to 9% among those with a definitive outcome.
  • Mean conversation length dropped 48-78% across different support categories.

The retention improvement from 66% to 70% is the most commercially significant number. Four percentage points of VPS retention at Hostinger’s scale translates directly into recurring revenue. Whether the fourfold signup growth is attributable to Kodee is less clear, and Hostinger does not claim it is.

For context, Intercom’s Fin, widely considered the industry benchmark for AI-powered customer support, reports an average resolution rate of 67% across more than 7,000 customers and 40 million conversations. Kodee’s 69% across all conversations is comparable – but Fin resolves support tickets and answers product questions. Kodee reads server logs, runs terminal commands, and deploys fixes on live infrastructure. Achieving a similar resolution rate on a fundamentally more complex task set is a different kind of result.

Why AI Retention Works Differently in Hosting

The retention problem is not limited to consumer apps. A ChartMogul analysis across 3,500 software companies found that AI-native companies post a gross retention rate of just 40%, compared to 82% net revenue retention for traditional B2B SaaS. RevenueCat’s data adds a nuance: AI apps actually convert users well, with trial-to-paid conversion 52% higher than non-AI (8.5% vs 5.6%). The problem comes after conversion. Users hit the limits of the AI, realize the output is not differentiated enough, or find that the novelty wears off.

Hosting reverses this dynamic. The product a customer is paying for – a VPS, a website, a server – exists independently of the AI. The AI removes the friction that causes churn in the first place: the configuration errors, the broken deployments, the 2 a.m. outage that the customer does not know how to debug. When AI is the product, users leave because the product disappoints. When AI is the support layer, users stay because the product becomes easier to keep.

GoDaddy Airo Takes the Other Path

GoDaddy, the largest domain registrar and one of the largest hosting providers globally, launched Airo.ai in November 2025 with six AI agents that handle business setup tasks: domain registration, website building, logo generation, compliance documents, and app development. The approach is fundamentally different from Hostinger’s. GoDaddy’s AI builds the customer’s online presence from scratch. Hostinger’s AI manages infrastructure the customer already has.

Both models use AI to reduce the barrier to entry. The open question is which model produces stronger long-term retention. Hostinger has published data suggesting its approach does. GoDaddy has not published comparable retention metrics for Airo.

The Question for the Rest of the Industry

Hostinger’s data is a single company’s self-reported numbers across one product line. It is not proof that AI agents universally improve hosting retention. But it is among the first publicly available datasets from a major hosting provider connecting AI deployment to measurable retention and acquisition outcomes. The industry has been talking about AI in hosting for two years. Hostinger is the first to publish what happened after shipping it at scale.

Whether other hosting providers see the same results will depend on their product, their customer base, and how deeply they integrate AI into the infrastructure layer rather than bolting it onto a help page. What Hostinger’s numbers do establish is a baseline: the conversation has moved from “should we add AI” to “what happens when we do.”