Valve Needed One Day to Beat Epic's AI-Coded Launcher

"Programming is hard," Epic wrote in its own launch announcement. "So, instead of doing that, we vibe coded a new Epic Games Launcher."

That is a real sentence from a company that makes a game engine for a living.

Epic spent years claiming it was the technically serious alternative to Steam, then let an AI write most of a new storefront, bragged about beating the incumbent by 46 percent, and got that claim functionally erased within a day by a competitor that didn't bother announcing anything.

On October 14, Epic shipped a preview of a rebuilt launcher using what it calls vibe coding—part of a "no handwritten code" policy. The claim: up to 46% faster startup than Steam, 5.47 seconds versus 10.07 seconds. The catch: Keywords Studio, contracted by Epic, averaged 5 tests on each of 13 machines and dropped the 3 slowest.

Translation: we paid someone to measure our homework, dropped the worst scores, and called the remaining average a scientific result.

The patch that didn't need a press release

Valve's response wasn't a blog post. It was a changelog line. A Steam Client update improved startup performance, hitting 5.23 seconds in testing, and arrived within 24 hours of Epic's victory lap.

That's the whole story. Epic hired a studio, ran machines, threw out the slow ones, and built a campaign around the result. Valve shipped a patch nobody was told to expect and made the comparison awkward.

PC Gamer's independent test found the Steam client at 9.32 seconds average, Epic at 5.98 seconds—Epic still faster on that rig even after Valve's patch. The post-patch picture is closer to a coin flip depending on your hardware.

The real problem for Epic: a company that bet storefront credibility on a clean, decisive speed win now argues over quarter-seconds against a decade-old client that didn't need a rewrite, didn't need an AI, and didn't need a press release to close most of the gap.

Vibe coding means nobody can explain the win either

Epic Games Store GM Martin Keely said the AI was essential: "We started from scratch. This is a completely new architecture... [AI] was a tool to write and review new features and new capabilities from the ground up in a different tech stack."

That's honest and also the problem. A storefront handles purchases, DRM, and account sessions. When code breaks in a traditional codebase, an engineer points at the commit. When code breaks in a vibe-coded one, the postmortem starts with "the model wrote this."

Valve won because a known, explainable codebase was easier to optimize in 24 hours than an AI-generated one was to out-build in months. If your pitch is engineering velocity, losing that argument to a patch note is the ballgame.

What to watch: whether Epic publishes an independent, Valve-free follow-up benchmark or lets the 46 percent number fade; and whether the vibe-coded launcher's bug reports start surfacing issues nobody at Epic can fully explain.