2 weeks ago we launched AgentCloak.
It's a free browser extension from @peteryared. It swaps your real data for similar fake data before it reaches ChatGPT, Claude or any other LLM.
Introducing AgentCloak: Use any AI without sharing your real data. Chinese AI services, ChatGPT, Claude, doesn't matter. You probably try to hide details before asking: different names, fake numbers, no address. But then the answer's useless because the AI is missing actual
Peter Yared @peteryared · Sep 18 · 1:57 · 367 · 1.2K · 7.1K · 2.2M
40 seconds into the video, we had Peter smash a Mac Mini with a baseball bat.
The results:
2.2M views
8,700 bookmarks. That's more saves than likes, which you rarely see.
3,000 downloads of the product
2,500 new followers for Peter
@BrianRoemmele and the former CPO of Tinder reposted it.
A few hours after launch, Edward Coristine found it on his own and quote posted it.
He's the government engineer who built the open-source model AgentCloak runs on.
Rampant from @ndstudio having real impact
Edward Coristine @as400495 · Sep 18 · quoting Peter Yared · 8 · 11 · 115 · 13K
The 2 million views are a nice number. A lot of launches get way more than that but don't convert or only achieve vanity.
The 3,000 users are what we're proud of. They mean the video, the strategy and the creators got people to actually try the product.
Here's how we did it.
The product
AgentCloak's parent company is seven years old.
For most of that time, it sold data isolation software to enterprises.
We were starting from 0 awareness on X.
Peter took Rampant, a small open-source model released by the US government's National Design Studio, and turned it into a browser extension.
It runs locally on your machine, making it essentially free to use.
The idea
I'll be honest about this as it's kinda funny. The concept we locked in on came out of the first call itself.
Peter spent 20 minutes ranting about the sovereign AI discourse.
Everyone on X was telling people to buy a $10K Mac Studio or stack H100s under their desk, just to keep their data private.
He believed that they're right about the problem and wrong about the fix.
We knew the structure of the video almost immediately:
Open on Peter's credentials
Rant about the hardware
Pivot: "you know what though, they're right about the problem"
Demo the product
Explain why it's free
Close
The script
My first draft went to Peter with this note:
Slack screenshot
This is how we write a lot of launch scripts now.
Every line starts at its spiciest version that we can still defend, with a tone-down alternate written inline.
Working script doc
On shoot day the founder records all of them. Then de-escalating becomes an edit decision instead of a reshoot or AI recreation.
We almost had to kill the smash
Peter almost talked us out of the Mac Mini.
He'd seen another founder video where someone breaks things in a glass cube, and he thought it looked silly.
His alternative: pull up apple.com, configure a $10K Mac Studio, see the price, and close the laptop in disappointment.
We shot both.
The smash stayed as the main moment. His idea became the backup ending.
We didn't need it. But having it is what let us go for the risky one.
He also asked for an AI-generated background instead of a real set.
We said no. A privacy product needs to look and feel real.
The shoot
We were meticulous on set.
Our crew captured every line in both versions, plus the backup ending.
We sat in video village the whole shoot, watching every take and calling for another when something felt off.
We created some of the AI scenes ourselves in post.
The smash
A Mac Mini is a solid block of aluminum.
Hit one with a bat and it dents and flies across the room in one piece. There's no debris and no money shot.
So we bought four dead ones on eBay for about $65 each and gutted them with our studio partner.
We scored the inside of the shells so they'd burst. Then we filled them with silver glitter, mylar confetti and PCB offcuts.
One of the comment on launch day which stayed with us:
We did not correct him.
The amplification
The video is maybe 30% of a launch.
The rest is ensuring its seen in the first few hours, or what good is all the effort we put into the video?
Step 1: Start with the buyer
Peter told us which enterprises he wanted to reach in healthcare, cybersecurity and other verticals.
These are companies where sending data to an LLM is a compliance problem.
We worked backwards from there. We looked at which influencers the execs at those companies follow, and built our list from that.
Step 2: Tier the influencers
Not all influencers do the same job.
We split them into tiers based on how big their following is and how relevant it is to Peter's buyers. Each tier posted at a different time.
Bigger creators went first. They quote posted and commented right after launch, which got the post to 200k views in the first 60 minutes.
Smaller influencers came next. Lots of them commented and reposted at once, so the post kept climbing after the first wave.
A few hundred accounts followed. Their reposts and likes kept the post moving through the day.
Step 3: Work on the copy
With a lot of the creators, we went back and forth on the hook and angle until each post sounded like them and still explained the product.
sir....we can't collect users' personal data ANYMORE they just launched a tool that swaps everything with FAKE DATA before it reaches us 😭
shirish @shiri_shh · Sep 18 · quoting Peter Yared · Paid partnership · 27 · 469 · 5.5K · 677K
This one quote post did 676k views.
We'd used the same format on an earlier launch, and just knew that we had to double down on a similar meme/angle.
Step 4: The founder's network
This is the highest leverage part of any launch.
If only influencers engage, the post stays inside the influencer echo chamber. It gets views but doesn't convert.
Peter's network is people who know him and trust him, and many work at the companies he sells to. When they engage, the post reaches their followers. That's where the signups come from.
What we did:
Had Peter DM every one of them individually over the two days before launch
Made each message personal, never copy-pasted
Asked for a quote post and a comment. The 2 highest virality signals to boost a post's visibility on X.
Timing
Most of Peter's network is on the East Coast, so we launched at 7am PT, which is 10am ET.
The launch was also on a Friday. That's not the normal Tuesday or Thursday most companies choose to launch.
The night before, Peter messaged asking if we should hold because a big AI release was eating the feed. We said go. Fridays are quieter, and a quieter feed you can dominate is better than a crowded one you can't.
Day one
Within a couple of hours, Peter's DMs were paying off.
His contacts were commenting and reposting. At 9:48, he messaged us:
"People I haven't talked to in years are pinging me."
One of the bigger creators liked the product so much that he asked to demo it on his livestream.
Creators get the post in front of the right people early. If the product is good, it spreads to people you could never pay to post.
Takeaways
The founder's first-call rant is more often-than-not the best launch video idea. Don't overthink it.
When scripting, write the unhinged version first. Tone it down in the edit.
Shoot every possible hook. You never know what will click until you trial & error.
Pick the audience first, then the influencers.
Tier your influencers.
Influencers are activation energy. As a founder, activating your own network is the highest-leverage thing you can do for your launch.
3,000+ signups. 2.2M views.
This is what we do at launchvideo.com.
We also bought launchvideos.com, with the 's'.
- Subah
