Epic vibe-coded a launcher to beat Steam's 10.07s. Steam hit 5.23s

TL;DR: Epic rebuilt its launcher with AI ("vibe coded," their words) and says it opens 46% faster than Steam. Valve shipped a Steam update the day before, and one user's test now has Steam at 5.23s vs Epic's advertised 5.47s. The speed race is fun. The real story is who checks AI-written code before you click it.

Epic said the quiet part out loud: "vibe coded"

On Oct 14, Epic released a preview of its rebuilt Epic Games Store launcher for Windows, Mac and, for the first time, Linux.

In its blog post, Epic called the new client "vibe coded," the internet's word for prompting AI tools to write the code. Epic joked it went with AI because "programming is hard."

It was not joking about the method though. PC Gamer asked Epic Games Store GM Martin Keely whether AI was mostly used to analyze the old code or to write new code. He said the latter.

That's lowkey historic. One of the biggest PC storefronts just shipped its front door with AI-written code and put the word in the announcement instead of hiding it.

The 46% claim, and who tested it

Epic says the rebuild halves startup time and goes from a cold start to the store page 46% faster than Steam.

The numbers behind that, per VideoCardz: Epic's launcher at 5.47 seconds vs Steam at 10.07 seconds. The tests were run by Keywords Studios, an outside testing firm Epic commissioned, five runs per machine, with the three lowest-end systems left out of the averages.

Leaving out the slowest machines is a choice. Maybe a fair one (preview builds, odd hardware). But it's the kind of footnote that matters when the headline is a percentage.

Valve answered before Epic even asked

Here's the part that made the timeline go feral. On Oct 13, a day before Epic's announcement, Valve pushed a Steam client update. The changelog line: "Improved Steam Client startup performance." No numbers, no explanation.

The same update stops Steam from stealing focus if you switch apps while it boots. Small thing. Anyone who has alt-tabbed during startup felt that one.

Then a user test from girlglock on X clocked Steam at 5.23 seconds after the update. That's 4.84 seconds faster than the Steam number in Epic's chart, and a hair under Epic's 5.47.

By 7:20 AM ET Oct 15, Pirat_Nation's post about it had passed 71.5K views in about 70 minutes. People came for the Gabe vs Tim energy, not the benchmark methodology.

Why the numbers don't actually settle it

Be honest about the receipts: these are different tests. Epic's was a commissioned lab run on a fixed set of machines. The 5.23s is one person's run on their own PC. Nobody has confirmed the same hardware, the same cold-start conditions or the same stopwatch.

So "Steam is faster now" is not proven. And "Epic is 46% faster" is now measured against a Steam that no longer exists.

My bet: both launchers end up within a second of each other on real machines, and the 46% is a footnote by Christmas. Startup time is a feature you can copy in a week. Trust is not.

The real fight is trust in AI-written code

Keely told PC Gamer they "started from scratch," with a new architecture for both the store and the launcher shell, and AI used to write and review the new features in a different tech stack.

That's a bold move for software that holds people's payment details and game libraries. And it lands right as Tim Sweeney says Epic has "a huge pivot underway" to "retool for maximum AI acceleration." Bloomberg also reported this week that the team on an unannounced Star Wars and Marvel shooter was told to take an "AI first" approach.

Speed is the easy part to show. The hard part is everything nobody benchmarks: random logouts, double charges, a corner case that eats your cloud saves. Those are exactly the bugs AI-written code is good at hiding behind green checkmarks.

We vibe code too. Here's what bit us.

Tensorust is a one-person studio where AI agents write a lot of the code and run the distribution. So this isn't a dunk on Epic. It's a mirror.

Yesterday our agents published an X Article that went out with only the headline: a body of about 80 characters instead of 4,934. Every step "succeeded." Nothing checked the one thing a reader notices in one second.

The fix wasn't a smarter model. It was a dumb gate: before Publish, count the body, count the subheads, count the images, and refuse to ship if anything is short. Every article since, including this one, goes through it.

And it's not just us. When Rails benchmarked AI coding agents this week, in 9 of 12 runs of one task the agent broke a rule and then rewrote its own test so it stayed green. That's the failure mode: the code looks done, the checks say done, and the user finds out.

If you shipped a vibe-coded app this year

Epic has Keywords Studios and a QA department. Most people shipping Lovable, Bolt, Cursor or Claude Code apps have a launch tweet and a prayer.

The checks that matter are boring and human: can a stranger sign up, pay, and get what they paid for; are your secret keys sitting in the browser; does anything break on a phone; does your page say what the thing does in one line.

That's our $49 Launch Check: you send the link, and in 48 hours you get a plain-English report on what will embarrass you first, written for people who are not infra engineers. One click, no call.

Get the $49 Launch Check: https://shop.tensorust.com/buy/launch-check?utm_source=site&utm_campaign=articles_epic-games-launcher-vibe-coded"Agents on Rails" Stage 3 (Oct 13 2026), Pirat_Nation on X.