The thing that made the agency good was research. Talha spent years working out which posts break
out and why, and for a long time that lived in his head, which meant it did not scale past him. So
the team codified it. They broke the process down step by step, taught a frontier model to run it,
and now that research happens continuously instead of whenever somebody had a free afternoon.
They build on Claude mostly, and switch when something better shows up. The tools
are the point, not the model underneath them.
In-house tool
Viral Edge
Watches roughly 15,000 reels a week across Instagram and scores each one on a
multiplier: how many times more views did this get than that account normally gets? A post at
400,000 views on a page that averages 20,000 is a signal. The same post on a page that averages
400,000 is nothing.
The outliers get shortlisted, the team reviews the best of them, and the ones that hold up get
written into the format library with the script structure, the shooting notes and the original
post it came from.
Why it wins: it separates a genuinely new idea from an account that is simply big.
In-house tool
Déjà View
One idea is usually worth several videos, and most people only ever make the first one. Déjà View
takes a piece of content and works out the other formats it could have been, drawing on a library of
over a hundred formats collected across the years.
The result is that a good idea gets several honest attempts at finding its audience, instead of one
shot and a shrug.
Why it wins: most content does not fail on the idea, it fails on the packaging.
In-house tool
Idea validation
Before a topic gets made, it gets checked. Live demand data from search, forums, comment sections
and competing posts, pulled in parallel and argued over, ending in a straight verdict on whether
the topic is worth anyone's week.
Why it wins: the cheapest edit is the video you decided not to shoot.
Knowledge system
A searchable memory for the agency
Every teardown, client lesson, meeting and format goes into one connected knowledge base of a few
thousand notes, linked so a lesson learned on a legal account can surface on a healthcare one.
Why it wins: agencies usually forget what they learned the moment the person who learned it is busy.
Production
AI inside the pipeline, not instead of it
Voice, illustration and parts of the edit are AI-assisted where it genuinely saves time. The
research, the calls and the person on camera are not.
The split: AI at the top of the funnel for volume, human taste at the bottom for selection.
The honest bit
Where he thinks AI is oversold
Talha is deliberately careful with AI clones and avatars. They are genuinely useful for a senior
executive who cannot get in front of a camera every week, and Dopameme builds them when that is the
real constraint.
But his view is that they take the personality out, and personality was the thing that was working.
Mixed in with a real person, they are a good tool. Used as a replacement for one, they quietly
remove the reason anybody was watching.
The position: use AI to do more of the work, not to be the person.