Happy Saturday. Elvis here.

Something I just realized recently - AI agents can amplify the same product launches by 1000x.

Most people just don’t realize this yet:

In my 8 years at Google, I launched products at Google I/O three times. Been at the Amphitheatre when the whole world is streaming. Also nuked 2M rows in a production table for an instant "hot-fix". (fun time, you should have seen how people reacted to this idea)

But you know what happens the week of launch? Code freeze.

For an entire week, nobody pushes anything to the codebase. Users complain something's broken? Come back later. A button is misplaced? Well, live with it.

Even with brilliant engineering teams, the fear of breaking something live is paralyzing.

In January this year I shipped ismypitchshit.com in 3 hours and did the exact opposite. I call it vibe-launching: iterating in real-time while the viral wave is still building.

Same thesis as the agent swarm essay that Karpathy called "brilliant or AI psychosis" — when you speed up the decision loop 1000x, everything changes.

The swarm essay was about coding. This one's about launching.

The Science of Virality

Most people think virality is luck. It's not. It's math — and it has only one number that matters: the viral coefficient (VC).

The viral coefficient measures how many new users each existing user brings in.

If VC is 1.1, it compounds exponentially. (It means 1 user will bring more than 1 other user) But if VC is 0.9, your product slowly dies.

The small difference between those two numbers is the difference between a product that fades and one that takes on a life of its own.

The Viral Loop

Basically here’s the idea I launched - user uploads a PR pitch and a journalist they are targeting. And we use AI to roast their pitch live.

So before I wrote a single line of code, I didn't just ask "will people use this?" I asked: "what makes someone compelled to share this the moment they see it?" And I built the answer into the product itself.

Every feature was a share trigger in disguise:

• Profanity in the domain — ismypitchshit.com. People share the URL just because it's funny. Stolen from motherfuckingwebsite.com.

• Turned the haters into users — Redditors hate self-promotion. PR people hate AI spam. This tool turns this psychology around, they were using the tool to prove they know better than the average spammers.

• Brutal AI is a pattern interrupt — Nobody wants “you are aboslutely right!”. When an AI calls your pitch "Absolute Garbage" that's shareable because it never happens elsewhere. (just like how 💩 emoji works great as a pattern interrupt in email subject line - you are here)

• Self-deprecation spreads — Wins look like bragging. Losses look relatable. People screenshotted their terrible scores.

• Sharing is built-in — easily screenshotable, every result is sharable with url.

But here's the thing: you can only plan so much before the idea hits your audience, after that it's new territory, which is why you need vibe-launching.

Vibe-Launching

The idea of vibe-launching is essentially to iterate on the product in real-time to push your viral coefficient beyond 1.0, aka going viral.

Here’s how.

After posting this on Reddit - I jumped straight into the session recording I’ve set up - so I can watched how real users behaved when they hit the site.

Because AI doesn't have our monkey brains that freeze under pressure — I could just tell it what to fix and it fixed it. I can never trust myself to code any of these on launch day.

Here's everything I saw and fixed—each within minutes:

Problem 1: People landed on the form on mobile, stared at it and left. Too many fields. Friction.
Fix: Added a "fill with random example" button. One click to see how it works. Instantly got more people through to step two.

Problem 2: On step 2 people still dropped off on the loading screen. AI was taking too long.
Fix: Added funny, on-brand loading messages. "Questioning your life choices..." "Checking if you actually Googled the journalist..." People stayed because they were entertained.

Problem 3: Now at step 3 people got their score but had no idea anyone else was using this.
Fix: Added a leaderboard, with timestamp and locations. Then let top scorers claim their spot with a name. Suddenly it was a competition, not just a tool. (This is exactly why Marc Lou's TrustMRR took off.)

Problem 4: Once leaderboard is in, conversion almost doubled, then I’m noticing that people didn't trust the score. Just a number felt arbitrary.
Fix: Added location to each score and a distribution graphic showing where they landed vs everyone else. Now it felt real with social proof.

Problem 5: Later I saw a Reddit comment that said scores were off.
Fix: Added a feedback mechanism. "Was this accurate?" Collected and displayed accuracy data in real-time.

Problem 6: Power users kept coming back — 3, 4, 5 submissions.
Fix: Built a hot lead capture after 3+ submissions. 20+ leads collected from this free tool.

Each fix took under 10 minutes to ship.

The product feedback loop just shrunk from weeks to minutes.

I watched someone get stuck and quit, figured out why, wrote the fix, pushed it to production, and watched the next user sail through — all while the Reddit thread was still climbing.

This is vibe-launching, the opposite of what I was used to at Google:

• Ship fast enough that "launch" is just the first commit

• Watch users in real-time (screen recording now, not dashboards in a day)

• Fix friction in real-time so virality compounds

• Add features based on actual behaviors, not roadmaps

The tool evolves faster than the virality fades. By the time the thread peaked, I'd shipped 6 updates and had the conversion path optimized down to each pixel.

The version that went viral wasn't the version I launched — it was something battle-tested.

The Real ROI (It's Not Revenue)

I didn't monetize this. No paywall. No upsell. No "premium roasts."

But it was the best marketing anyone could have asked for:

1. Built an "anti-villain" brand.

My main product helps PR people pitch journalists. The biggest objection? "AI pitches are spam. You're part of the problem."

Now I have a tool that publicly roasts bad pitches. I'm not pro-spam. I'm the guy who built the thing that destroys spam.

Brand positioning solved forever. (when I showed this on a sales call client got so excited and asked “that was you?!”)

2. Created a permanent proof point.

Every time someone says "I don't trust AI to do this" — I point them here.

It’s a brand asset that compounds forever:

800+ pitches graded. 82% garbage tier (the tool was honest). Reddit upvotes. Comments from actual journalists saying it works. Accuracy graph displayed. Hundreds of PR pros spreading it voluntarily.

That's not a demo. That's a courtroom exhibit. (Zoe wrote this line - AI-sounding but real good lol, more about her below if you don’t know her yet)

The Takeaway

1. AI changes how I think about launches.

Code freeze made sense when shipping was risky and slow. Now you can watch users in real-time and fix things in minutes. Don't freeze when it goes viral — that's when you should be shipping fastest.

2. Create viral moments that become brand assets.

Most founders chase virality for revenue. The smarter play is building permanent proof points. A viral tool that validates your expertise compounds forever. That's what PR does and marketing can’t do.

—Elvis

P.S. I'm building Medialyst with an AI cofounder named Zoe. We've somehow grown from 6k to 42k followers on X since my last newsletter — just by documenting everything we do along the way. The agent swarms, the 2am production fire she made, and the vibe-launches. Follow along here: thread