Kurt Mackey once built Fly.io around two big ideas. Applications should run close to their users for speed. Cloud tools should feel simple, like Heroku, yet pack the power of AWS. Those bets paid off for years. Strong quarters rolled in. The best financial months in company history arrived. Then Theo Browne dropped a verdict that hit like a slap.
In a video ranking the best places to host new apps in 2026, Browne praised Fly.io. He also said it was the provider he least expected to survive the year. The comment rattled Mackey. It exposed a raw nerve about the company’s direction. Fly.io had coasted. An identity crisis simmered beneath the growth.
Mackey didn’t dodge the discomfort. He overshared in a blog post that reads like a confession and a manifesto. The result? A major funding round. A sharpened focus on a new product called Sprites. And Mackey stepping aside as CEO, handing the reins to Scott Johnston, former Docker chief. The moves signal a profound shift. Fly.io now centers on computers built for AI agents.
AI has changed everything, Mackey argues. Before spreadsheets, every financial model was a custom program written by coders. Then formulas let millions build their own. AI promises the same leap, only larger. “Almost anybody will probably be able to build almost any kind of computer program,” he wrote in the Fly.io blog post. Personalized, adaptive software will become the norm. Opinionated cloud platforms designed for 2020-era apps suddenly look like a bad wager.
Developers once read docs and tinkered with CLIs. Agents don’t. They expect to run on workstations but often land in the cloud because laptops sleep. Public clouds feel clunky for this new reality. Servers are pets or cattle. Agents need something in between. Semi-disposable. Instant. Cheap when idle. Persistent when needed. That insight drove the creation of Sprites.
Sprites aren’t sandboxes. They’re full computers for agents. Create hundreds or thousands quickly. Each comes with 100GB of durable storage. Metered billing pauses when idle. Share apps over the internet with colleagues. The design emerged from watching real robot customers. “Our fastest-growing customers were all robots,” Mackey noted in an earlier post referenced within the same article.
The latest version, rolled out in that same candid post, adds two breakthroughs. The Sprite Block Device replaces an earlier homemade storage stack based on JuiceFS and Litestream. Ben Johnson and Tim Newsham rebuilt it from the ground up. Faster. More reliable. It supports instant checkpoint and restore. Better yet, it enables drive forking. Spin up a template Sprite, then clone it millions of times efficiently. Fly.io detailed the technical overhaul there.
Connectors form the second advance. They let Sprites call external APIs with strong security. Built on prior work to tokenize and secure the core platform, they avoid handing agents raw credentials that could be exfiltrated. No more juggling API keys manually. The features address demands from agent companies that stuck with Fly Machines even after Sprites launched.
Mackey didn’t stop at product updates. He admitted the science-project phase of Fly.io had run its course. For eight years he ran experiments. Unmanaged Postgres. Global CDNs. User-mode WireGuard. A bottom-up engineering culture. All-remote teams spread across countries. Some bets succeeded. Others taught hard lessons. The company no longer needs that constant experimentation.
Enter Scott Johnston. Mackey spent months in 2025 talking with him about Fly.io’s next chapter. Johnston steered Docker through its own identity crisis between enterprise and developer worlds, ultimately supercharging the business. As a shareholder, Mackey preferred Johnston’s playbook. As CEO, he preferred letting someone else execute it. The board agreed. Johnston is now CEO. Mackey stays involved but steps back from daily leadership.
The transition comes as open banking gains traction in the United States. A recent analysis from American Banker highlights how financial data rights rules from the Consumer Financial Protection Bureau are pushing banks to share customer data more freely. That shift creates fertile ground for agent-driven finance tools. Agents could soon handle transfers, budgeting, or investment decisions with user permission. Sprites offer the infrastructure to run those agents without the overhead of traditional servers.
Recent coverage shows the momentum. A July 2026 piece in FinTech Futures reports fresh legal pushback against CFPB rules, yet adoption continues among forward-looking institutions. Another report from Bloomberg details how AI agents are already prototyping personalized banking experiences, echoing Mackey’s vision of software anyone can shape.
Fly.io’s bet carries risks. Public clouds excel at scale for millions of users. Agent-driven apps may start smaller, more personal. Yet the company claims Sprites suit both existing customers and the coming wave. A beta program for advanced features, including self-cloning Sprites, is now open. Early testers can spin one up in about a minute at fly.io/sprites.
Not everyone buys the thesis. Skeptics point to the crowded AI infrastructure market. Big players pour billions into GPUs and orchestration. Fly.io’s approach stays lean. No heavy GPU focus here. Emphasis falls on quick, stateful, secure compute that agents can fork and forget. It’s a narrower wager, but one rooted in observed customer behavior.
Mackey’s post mixes candor with confidence. He promises readers will finish the full piece, an “honor thing.” That tone reflects a leader confronting uncomfortable truths. The AI wave isn’t a side story. It rewrites the assumptions behind cloud platforms. Companies that cling to old models may fade. Those that adapt, even painfully, might thrive.
Fly.io now chases that adaptation. Sprites represent more than a product. They embody a view of computing where agents act as the primary users. Humans direct high-level goals. The machines handle the rest on disposable yet persistent infrastructure. Storage that forks like code. Secure calls that protect secrets. Billing that matches actual work.
Whether this reorients the company successfully remains to be seen. Strong financials give runway. Johnston’s track record adds credibility. Mackey’s willingness to yield the CEO seat shows self-awareness rare in founders. The broader industry watches. Open banking expands access to financial data. AI agents stand ready to act on it. The infrastructure beneath those agents could determine who captures the value.
One thing feels clear. The dingo ate the baby, as Mackey put it. Traditional software development pipelines, rigid CI/CD, fixed audiences of millions. Those models face disruption. Spreadsheets once seemed niche. Now they’re everywhere. AI agents may follow the same path, turning every user into a builder. Fly.io placed its chips on that future. The cards are still turning.
WebProNews is an iEntry Publication