<- all tokdocs
2026.09.25@aistartupfren

Diogo Almeida started launch day with 1,500 followers and ended with 39.5M views on one post

Marketing
Watch on TikTok

View on TikTok ->

The audio in this 80-second TikTok says "Jeff the AI model" and "Doomers AI"; the frames show the model is called Jev and the agency is called Doomers. Uploaded 2026-09-25 by @aistartupfren under the title "founders can learn from Jev launch success #aistartup #aiagent #start...", the clip had 3,088 views, 177 likes, 12 comments, 23 reposts and 127 saves at capture. Forty frames were extracted at two-second intervals. Frames 1 through 3 are a talking head over a beige wall with the caption card "Jev AI is a huge marketing success". From frame 4 onward the background becomes a screen recording of a single web page, breadcrumbed "DOOMERS / CASE STUDIES / TYPESAFE AI CASE STUDY" and headlined "TypeSafe AI and Jev: from stealth to the largest launch we have run". Two stat tiles are legible behind the speaker's head: "39.4M / VIEWS ON THE PO[ST]" and "142K / FOUNDER F[OLLOWERS]". The recording scrolls through three sections. One is captioned "What the post led with" and highlights the sentence "The post led with the one line a stranger can verify: the founder co-created ChatGPT. Then the question the company exists to answer, why superhuman chat models have not led to AGI." Below it, partially covered by the speaker's face, sits "The video runs just under three minutes and [was] made by Mark Vassilevskiy at Skale (@MarkKn...)" and a red hyperlink reading "launch videos over two minutes do twice the median of videos under thirty seconds". The other section is captioned "Distribution and seeding" and highlights "Doomers seeded the announcement to between 80 and 100 specific accounts: engineers, founders, builders and creators whose audiences overlap the people TypeSafe needs to reach: researchers, engineers, and the operators [who decide] what their teams try next." None of the names on that page, not Jev, not TypeSafe, not Mark Vassilevskiy, not Skale, are ever spoken aloud.

Whisper turned one company and one model into two different words

The transcript renders the model name as "jeff" six times:

the success of jeff the ai model is a carefully orchestrated marketing success

The on-screen title card in frames 1 through 3 reads "Jev AI is a huge marketing success", and the TikTok description carries the hashtag "#jev". The agency is transcribed as "doomers ai"; the page header in every frame from 5 onward reads simply "DOOMERS", and the site lives at doomers.ai.

One more line is mangled past recovery:

based on access for you page algorithm

The launch ran on X, and the Doomers case study is explicit that the seeded accounts were reached "in the window the ranker" evaluates. The likely intended phrase is X's For You page algorithm. Whisper heard "access" where the speaker said the letter X.

The product itself is real and recent. Jev is the first model from TypeSafe AI, released 15 September 2026. Its founder and CEO is Diogo Almeida, who spent roughly four years at OpenAI working on RLHF, InstructGPT, ChatGPT and GPT-4 before leaving in 2024. The Register reported the company had raised $40 million. So the speaker's compressed claim that the post "led with the founder who co-created ChatGPT at OpenAI" matches both the Doomers text on screen and Almeida's own launch post, which opened "After co-inventing ChatGPT, I kept asking myself: why have superhuman chat models not led to AGI?"

The seeding number is 80 to 100, and the deciding window is one hour

The audio states the number twice as a round hundred and attaches a day-long deadline:

seeding this announcement this viral post amongst a hundred specific related accounts basically asking a hundred content creators who are engineers founders builders to post about the launch within 24 hours

The sentence highlighted on screen in frames 23 through 40 gives a range, not a round number:

Doomers seeded the announcement to between 80 and 100 specific accounts: engineers, founders, builders and creators

The 24-hour figure does not come from the seeding paragraph. The line directly above it, visible in the same frames, sets a much shorter clock:

A launch earns almost all of its reach from people who did not follow the account that morning, and it earns it or fails to inside the first hour.

The case study also does not describe the 80 to 100 accounts as being asked to post. It describes them as accounts "whose audiences overlap the people TypeSafe needs to reach", reached inside the ranker's window. The paragraph that follows, which the speaker paraphrases as a snowball effect, reads "From there the launch travelled on its own. It was picked up organically around the world, by researchers, press and communities nobody had contacted." That part of the video holds up.

150,000 signups were waitlist entries, and the counters on screen were already stale

The audio pairs two numbers:

they did two main things to get 39 million views as well as 150 000 users in the first 24 hours

The Doomers case study labels the second figure as "150K+ waitlist signups in first 24 hours". A waitlist entry and a user are different things, and the case study never claims 150,000 people used Jev on day one. TechCrunch reported the related symptom, that TypeSafe "briefly lost the ability to serve users from its API because demand was so high".

The first figure is a live counter, and the frames caught it mid-climb. The screen recording shows 39.4M views and 142K founder followers. Reading the same page today returns 39.5M views, 144K founder followers and 99.9K company followers. Earlier indexed snapshots of Doomers pages carried 37.2M and 35M+. The creator screen-recorded the page at one moment and narrated a rounded-down "39 million".

Two other numbers sit on that page. The case study ranks the launch "4 of 699 launches" tracked by Doomers; the separate Doomers launches entry for TypeSafe AI ranks it "#4 of 706". The denominator moves because Doomers keeps adding launches to the directory. That entry also lists 76K likes, 8K reposts, 4K replies, 7K quotes and 59K bookmarks, and compares the post against a product-news category median of 206K views.

Doomers also records where Almeida started. The case study gives the founder's opening follower count as 1,500. The gap between 1,500 followers and 39.5M views on a single post is the actual claim buried under the video's rounded numbers, and the speaker never mentions it.

The two-minute video claim is real, and it has a measured number behind it

This is the part of the video that survives verification most cleanly. The audio:

counterintuitively they made a video that's over two minutes long apparently a launch video that's over two minutes long is better than a 30 second long one because people want to understand how your product works

The hyperlinked red text in frames 20 through 22 states it with a multiplier:

launch videos over two minutes do twice the median of videos under thirty seconds, because a demo that shows [the] model doing the whole job holds attention a[nd] a ranker that grades on watch time rewards it.

The video in question runs 2 minutes 56 seconds and was made by Mark Vassilevskiy at Skale, credited on screen as "Mark Vassilevskiy at Skale (@MarkKn...)". His X handle is @MarkKnd and his LinkedIn lists him as CEO of Skale.solutions. The speaker's reasoning ("people want to understand how your product works") is a looser version of the page's reasoning, which is about watch time as a ranking input rather than about comprehension.

The three claims the post led with disagree with TypeSafe's own blog on one number

Behind the speaker in frames 13 through 21, a partially obscured paragraph reads "then three claims a reader can [check in] the video: 20 to 200 times faster, [40 to 400] times cheaper, output tokens fre[e]". Almeida's launch post used the same three: "20-200x faster", "40-400x" cheaper, output tokens free.

TypeSafe's own launch blog, "Introducing System One Models & Jev", puts the speed range differently. It describes Jev as "40x-200x faster for the same levels of frontier intelligence for System One shaped queries", with end-to-end response times of 70ms to 500ms against 3 to 329 seconds for frontier LLMs. The cost side is consistent across both: input tokens at "$0.042 / MTok ($42 per billion tokens)" and output tokens "FREE (too cheap to meter)". The blog's own workflow evaluation cites a specific pair, "193.6x faster, 444.6x cheaper".

So the floor of the speed claim differs between the marketing post (20x) and the engineering blog (40x). Independent testing reported by TechCrunch landed lower than both: a Vercel engineer measured Jev returning results five to 18 times more quickly than OpenAI's Luna model.

Key Takeaways

  • Whisper wrote the model name as "jeff" throughout and never once produced "Jev"; the correct spelling appears only in the frames, in the caption card "Jev AI is a huge marketing success", and in the video's own hashtags. The transcript line "based on access for you page algorithm" is a mishearing of a reference to X's For You page.
  • Every proper noun that makes the story checkable exists only in the frames. TypeSafe AI, Jev, Mark Vassilevskiy, Skale and the URL doomers.ai are all visible on screen and none are spoken.
  • The video rounds 80 to 100 seeded accounts up to "a hundred content creators" and attaches a 24-hour posting deadline that the case study does not make. The case study's stated window is the first hour.
  • The 150,000 figure is waitlist signups in the first 24 hours, which the video calls "users".
  • The stat tiles in the frames read 39.4M views and 142K founder followers. The same page now reads 39.5M and 144K. The creator screen-recorded a counter that is still moving.
  • The two-minute video claim checks out and is more specific than the video says: over two minutes does twice the median of under thirty seconds, and the Jev demo runs 2:56.
  • Unverified: I could not confirm the speaker's disclosure "they didn't pay me", since no disclosure register exists for the 80 to 100 seeded accounts and Doomers does not publish the list. I could not confirm whether @aistartupfren was among the seeded accounts or picked the launch up organically ten days later. I could not read the obscured portions of the case study behind the speaker's face, including the full "From there" paragraph tail and the "six stages" line that appears cut off at the bottom of frames 23 through 40, so I verified those passages against the live page rather than the frames. The 20-200x versus 40-200x speed discrepancy is documented on both sources but TypeSafe has not published a reconciliation of the two ranges.

Resources

Published September 25, 2026. Writeup generated from a favorited TikTok.