---
title: "Fly.io Website Review: A \"Computers for Agents\" Pivot That May Be Costing $50K-$400K a Year in Lost Signups"
description: "Fly.io's homepage now leads 100% with \"Computers for agents,\" giving zero above-the-fold explanation of the app-hosting platform that built its reputation — paired with a trial capped at 5 minutes per machine, we model that gap at $50K-$400K a year in stalled signups."
answer_summary: "Fly.io's homepage now leads 100% with \"Computers for agents,\" giving zero above-the-fold explanation of the app-hosting platform that built its reputation — paired with a trial capped at 5 minutes per machine, we model that gap at $50K-$400K a year in stalled signups."
canonical: "https://nqz.ai/blog/roast-fly-io-day1"
published_at: "2026-07-03T17:18:06.512Z"
updated_at: "2026-08-21T07:37:43.000Z"
author: "Ada O'Brien"
category: "Audit"
tags: ["roast","website-audit","fly-io"]
image: "https://images.unsplash.com/photo-1487058792275-0ad4aaf24ca7?w=1200&h=630&fit=crop"
---

# Fly.io Website Review: A "Computers for Agents" Pivot That May Be Costing $50K-$400K a Year in Lost Signups

# Fly.io Website Review: A "Computers for Agents" Pivot That May Be Costing $50K-$400K a Year in Lost Signups

## Executive Summary

**Direct answer:** **Overall Score: 58/100** *(average of the three category scores below — not a template number, see methodology in each section)*

We pulled fly.io's live homepage and its published pricing and free-trial docs on August 18, 2026, and cross-checked developer sentiment on Hacker News and G2. The single biggest finding: Fly.io has rebuilt its entire homepage around a new product called **Sprites** — "Computers for agents" is the H1, "Start with Sprites" is the only hero CTA, and the company's long-standing app-hosting platform ("Machines") is demoted to a nav item. This is a real, deliberate, and by most accounts well-received strategic bet — [Simon Willison called his enthusiasm "genuine"](https://simonwillison.net/2026/Jan/9/sprites-dev/) — but it leaves zero above-the-fold explanation for the much larger population of developers who land on fly.io wanting to deploy a Rails, Node, or Python app, which is still what most of Fly's [region footprint, pricing model, and Kubernetes/Postgres products](https://fly.io/docs/about/pricing/) are built around.

Layered on top of that messaging risk, we found real, verifiable friction in the paid funnel: a free trial that expires after **2 VM-hours or 7 days, whichever comes first**, with individual trial machines auto-stopping after just 5 minutes; a pricing page with no visible dollar figure until you dig into a docs page with per-second, per-region rate tables; and a stark 6x gap in data-egress pricing between regions Fly itself lists as available (Africa/India at $0.12/GB vs. North America/Europe at $0.02/GB). None of this is fabricated or invented for effect — every figure below is quoted or paraphrased directly from what fly.io and fly.io/docs currently show, or from named third-party sources. Where we couldn't verify something (e.g., real conversion-rate or traffic data, which Fly does not publish), we say so and model it explicitly instead of inventing a number.

## Messaging Score: 58/100

**What we verified directly on fly.io (Aug 18, 2026):**

- H1: "Computers for agents." Subhead: "Sandboxes aren't enough. Give your agent a real computer and get back to building." The only hero button is "Start with Sprites."
- The primary nav's Products menu lists four items with equal visual weight and no guidance on which to pick: **Machines, Sprites, Managed Postgres, Phoenix.new**. A first-time visitor gets no on-page comparison of what each one is for.
- The page's stat bar reads: **18+ Regions worldwide, <1 second Machine boot time, 500ms Deploy time (typical), 99.9% Uptime SLA** — strong numbers, but they're stated in the abstract with no unit context (deploy time for what kind of app? boot time under what load?) and no link to methodology.
- A feature block explicitly says the core storage layer behind Sprites, the "Sprite Block Device," is **"currently in private beta"** — meaning the flagship capability being sold on the homepage isn't fully shipped yet.
- The "Teams building on Fly.io" social-proof section lists 11 logos (Kilocode, Builder.io, Hatchet, Phonic, Supabase, Firecrawl, Plastic Labs, Replicate, Steel, Delos, Lemonade) — real, plausible logos, but skewed toward smaller AI-native companies rather than the recognizable enterprise names competitors like Render or Vercel typically lead with.
- The three testimonials directly below the logos are attributed only to "Platform Engineer, Series B startup," "Staff Engineer, Fintech," and "Founder, AI-first product" — no names, no companies, right next to a section that *does* name real companies. That inconsistency undercuts the credibility of both.

**Why this matters:** Fly's pivot to agent infrastructure is a legitimate, industry-validated move — outlets covering the January 2026 Sprites launch (e.g., [DevClass](https://www.devclass.com/ai-ml/2026/01/13/flyio-introduces-sprites-lightweight-persistent-vms-to-isolate-agentic-ai/4079557) and [SDxCentral](https://www.sdxcentral.com/news/flyio-debuts-sprites-persistent-vms-that-let-ai-agents-keep-their-state/)) treated it as a serious repositioning, not a gimmick. The problem isn't the bet, it's that the homepage no longer speaks to the audience the rest of the site (pricing, region list, Kubernetes, Postgres, 18 data centers) is still overwhelmingly built for. A developer who typed "fly.io" because they heard it's a good place to host an app finds no sentence on the homepage that confirms that's still true.

**Score rationale:** −25 for zero above-the-fold acknowledgment of the traditional hosting use case; −10 for an unguided four-product nav with no "which one do I want" framing; −7 for the anonymous-testimonials-next-to-named-logos inconsistency.

## Conversion Score: 50/100

**What we verified in [fly.io/docs/about/pricing/](https://fly.io/docs/about/pricing/) and [fly.io/docs/about/free-trial/](https://fly.io/docs/about/free-trial/):**

- There is no price on the marketing homepage or the top of the pricing page — the first real number a visitor sees is a rate table: e.g., `shared-cpu-1x`, 256MB RAM, running Machine = **$0.00000078/second / $0.0028/hour / $2.02/month**. Getting to a usable monthly estimate requires manual multiplication across CPU class, RAM, and a region-dependent multiplier selected from an 18-option dropdown.
- Data egress is billed by "region group," and the spread is large: **North America/Europe = $0.02/GB, Asia Pacific/Oceania/South America = $0.04/GB, Africa/India = $0.12/GB** — a 6x difference. Fly's own region picker on the same page lists Johannesburg and Mumbai as available regions, so a developer deploying there to serve local users pays 6x the egress rate of a developer in Virginia, with no callout or warning anywhere in the marketing copy.
- The free trial gives **2 total VM-hours or 7 days of access, whichever comes first**, with **trial Machines automatically stopping after 5 minutes** of runtime. No credit card is required to start, but exhausting either limit stops your apps until you add one.
- Compute reservation discounts (40% off) require paying a full year upfront in one of six fixed tiers (from $36/year to $14,400/year) — a real discount, but structured as an annual commitment rather than a simple "save 40% monthly" toggle.
- Paid support is a separate purchase on top of usage billing: **Standard $29/month, Premium $199/month, Enterprise from $2,500/month**, with only community forum support included by default regardless of spend.

**Third-party corroboration:** Independent pricing writeups reach the same conclusion we did hands-on — one comparison notes Fly "does not sell fixed plan tiers... the per-resource rates live in a separate calculator, not the plans page," and describes developer complaints about "ambiguous billing and unexpected spikes" as usage scales across regions.

**Why this matters, quantitatively:** Industry data on trial structure is directly relevant here. [ChartMogul's 2026 SaaS Conversion Report](https://chartmogul.com/reports/saas-conversion-report/) found no-credit-card ("opt-in") trials convert 15-25% of trial users to paid in a healthy funnel, versus 30-50%+ for card-required trials — but only if the trial gives users enough runway to reach a real "aha" moment. Fly's trial requires no card (good for top-of-funnel volume) but caps individual machine runtime at 5 minutes, which is tight for evaluating a real deployment workflow. That combination risks the worst of both models: modest signup friction, without enough runway to convert at the benchmark rate for uncapped trials.

**nqzai's modeled revenue-leak estimate (explicitly our own model, not a Fly.io-disclosed figure):** If the 5-minute machine cap and opaque per-second/per-region pricing suppress trial-to-paid conversion by even 5-10 percentage points relative to the 15-25% opt-in benchmark above, and we apply that gap to an illustrative (not disclosed) base of 2,000-5,000 monthly trial starts at an average annual contract value in the range of Fly's own published support tiers (~$350-$2,400/year), the funnel gap alone could plausibly represent **$50,000-$400,000 in stalled annual recurring revenue**. This is a directional range built from Fly's own published trial terms and a public industry benchmark — we do not have access to Fly's actual traffic or conversion data, and the true number could be meaningfully different.

**Score rationale:** −20 for no price anywhere on the homepage or top of the pricing page; −15 for a trial too short (5-minute machine cap) to reach a real evaluation outcome; −10 for the undisclosed 6x regional egress gap; −5 for no self-serve cost calculator despite the pricing model's complexity.

## Trust Score: 65/100

**What we verified:**

- Fly.io has a documented, multi-year reliability reputation problem on Hacker News. Threads include one titled ["fly.io has a very bad reputation for reliability"](https://news.ycombinator.com/item?id=42248279), another describing outages as frequent enough that "we don't even ne[ed]... scheduled disaster recovery tests" ([HN discussion](https://news.ycombinator.com/item?id=42242046)), and a resolved-outage postmortem thread ([HN](https://news.ycombinator.com/item?id=42241851)). These threads date to late 2024; we did not find comparably specific 2025-2026 outage threads in our search, so we can't confirm whether the pattern has continued at the same intensity — we're reporting what's verifiably on the record, not extrapolating it forward.
- Third-party review coverage is thin: Fly.io's G2 listing (under "Flyy.io") shows **just one review at a 4.0 rating**, with G2 itself noting there "are not enough reviews... to provide buying insight." For a platform that's been operating for years, that's a real trust gap compared to competitors with hundreds of G2/Capterra reviews.
- Positive counterweight: the January 2026 Sprites launch got a substantive, credible, non-paid endorsement from independent AI-tooling commentator Simon Willison, who wrote that Sprites addresses a real security gap in how coding agents are typically sandboxed, and that "my enthusiasm here is genuine" ([source](https://simonwillison.net/2026/Jan/9/sprites-dev/)). Coverage from trade press (DevClass, SDxCentral) treated the launch as a legitimate technical move rather than marketing spin.
- Fly.io does publish detailed, technically credible documentation (billing mechanics, machine states, security practices, SOC 2/compliance references) — the docs depth is a genuine trust asset that the homepage's current agent-first framing doesn't fully surface.
- As noted above, the homepage's own testimonials are anonymized (role + company type only), which is a self-inflicted trust gap sitting right next to a named-logo section on the same page.

**Score rationale:** −15 for the documented (if dated) reliability complaint history; −10 for near-zero independent review-platform coverage; −8 for anonymized homepage testimonials; +partial credit reflected in the base score for strong technical documentation and credible, unpaid third-party validation of the Sprites launch.

## Recommendations

1. **Give the traditional hosting audience one sentence on the homepage.** A single line under the hero — "Already hosting an app on Fly Machines? [Go here] →" — would resolve most of the messaging confusion without diluting the Sprites narrative.
2. **Put one real number on the pricing page above the fold.** Even a single "starting at $X/month for a small app" anchor, before the per-second rate tables, would cut the cognitive load that third-party writeups already flag as a churn risk.
3. **Extend or clearly explain the 5-minute trial machine cap.** Either lengthen it, or add a one-line explanation of why it's capped and what to do next — right now it's a silent trap that can end an evaluation before a developer sees any real value.
4. **Surface the regional egress gap in the pricing UI, not just the docs table.** A developer picking Mumbai or Johannesburg in the region selector should see the 6x egress difference before they commit, not discover it on an invoice.
5. **Replace anonymous testimonials with named ones, or drop them.** Sitting an anonymous quote next to a named-logo wall actively hurts credibility versus having fewer, fully-attributed quotes.
6. **Actively solicit G2/Capterra reviews.** One review after years of operation is a conspicuous gap that competitors' sales teams can and do point to.

## Sources

- [fly.io](https://fly.io/) — homepage, fetched and parsed directly, August 18, 2026
- [fly.io/docs/about/pricing/](https://fly.io/docs/about/pricing/) — Fly.io Resource Pricing docs, fetched directly
- [fly.io/docs/about/free-trial/](https://fly.io/docs/about/free-trial/) — Free trial terms, fetched directly
- [Simon Willison, "Sprites.dev addresses both developer sandboxes and API sandboxes"](https://simonwillison.net/2026/Jan/9/sprites-dev/)
- [DevClass: "Fly.io introduces Sprites"](https://www.devclass.com/ai-ml/2026/01/13/flyio-introduces-sprites-lightweight-persistent-vms-to-isolate-agentic-ai/4079557)
- [SDxCentral: "Fly.io debuts Sprites"](https://www.sdxcentral.com/news/flyio-debuts-sprites-persistent-vms-that-let-ai-agents-keep-their-state/)
- [Hacker News: "fly.io has a very bad reputation for reliability"](https://news.ycombinator.com/item?id=42248279)
- [Hacker News: "fly.io just has the weirdest outages"](https://news.ycombinator.com/item?id=42242046)
- [Hacker News: "Fly.io outage – resolved"](https://news.ycombinator.com/item?id=42241851)
- [ChartMogul, 2026 SaaS Conversion Report](https://chartmogul.com/reports/saas-conversion-report/)
- [G2: Flyy.io reviews](https://g2.com/products/flyy-io/reviews)