I went into psychosis and ended up solving marketing research

I went into full psychosis mode over the past few weeks. I was (way too) deep in Claude Code with many terminals open at once, connecting everything & building agents to solve internal problems where our custom data is an edge.

It started with Clawdbot and ended up leading to something much bigger: solving the messy, complex, and time consuming problem of marketing research.

You do not “automate” marketing.

You cannot even automate a question like “What strategy should I use for app X?” as a one shot answer. Why?

First, because there is just too much context. A real good strategy has many proprietary details.

Second, and more importantly, marketing moves insanely fast. A hook only has a limited amount of attention it can capture across multiple uses.

Once the people who were meant to see it have seen it enough times, it loses its power. Even the more evergreen ones, like “Why tf is [language] so hard,” are starting to lose momentum.

And the biggest wins usually come from original strategy creation, especially around formats and hooks (right behind that comes fast competitor cloning).

A strategy is not built all at once in its final form. It takes shape over time, layer by layer, as you execute, learn, and improve through the feedback loop.

So you cannot automate it + you also cannot just copy whatever is already working, because too many people will do the same and the upside quickly collapses for everyone.

That is why I’d say we’re “solving marketing” as a value proposition carefully.

An AI that tells you EXACTLY WHAT TO DO is a fallacy. If it was even possible, on top of removing you from the equation anyway, it would completely remove any advantage once everyone uses that AI (then only two things would still matter: strategies where capital is an edge, and original strategies).

What can be partially solved though is research and monitoring, and that matters a lot. .

The best strategies usually come from studying what others are doing, then iterating to create a better version. That is because more than 50% of a growth marketer’s job is finding and processing the right data and information.

But research isn’t solvable by processing ALL the data. That just injects massive amounts of noise. TikTok and Reels are full of noise. Only a small subset of videos and accounts actually matter.

My belief is simple: the right corpus of data matters more than ALL the data. And Social Growth Engineers has spent more than a year handpicking that corpus.

Hook Intelligence

Hook intelligence may be the single most important layer of all. We have said this again and again: if a hook works once, it will probably work multiple times.

And if the most important part of a video is the first 1 to 2 seconds that lock in retention, then on-demand hook intelligence is exactly what you want. And you want it every single day, because these things get recycled nonstop.

Strategy Breakdowns

Take an app with 200+ accounts. It takes us hours to go through everything, review the videos, group the format strategies, and pull out the real details and growth hacks. There is too much detail, too much context, and too much noise for models to handle that cleanly on their own.

But when we do the heavy lifting first and feed those breakdowns into an LLM, the model is grounded in real facts and can use those facts intelligently.

Trends & Format Ideas

Trends and format ideas can also be separated from the layers above. A format idea is usually spotted by a growth marketer with the right eye for it, someone who understands a vertical deeply enough to see that a format could work in another way or another context.

And trust me, this is where the biggest wins can come from. It was during our period tracking deep dive that we identified the formats that later drove hundreds of millions of views for Flo.

And the alarm format we identified and shared literally helped Evry and Wayk each cross $50K a month in less than 60 days.

Tying all together

At some point, I realized we were collecting and refining an absurd amount of valuable data through all this heavy lifting, writing 10+ articles a day and covering 10 new apps a day, but a huge part of that curated value stayed trapped unless the right person happened to find it.

So I plugged everything into Clawdbot. And it instantly clicked. Suddenly I could query everything, find articles and apps, and pull growth hacks exactly when I needed them.

What it still needed was more processing on the data side so the LLM could access better tools. That is when Prometheus was born.

Prometheus has been our logo for a while, and the name fits. Prometheus brings fire and knowledge from the heavens down to humans, and gets punished for giving away the sauce.

SGE’s Prometheus does the same thing. It accelerates research by 1000x — works directly through Slack with our native agent and wherever your Clawdbot lives, and gives you the videos and information you need without making you jump between pages.

That means intelligence can spread really fast across the marketers and creators inside your company. The edge becomes massive. It is already fully rolled out internally across all our creator managers and strategists.

Try it with your premium subscription here: socialgrowthengineers.com/prometheus

I built this to help my app team reach a billion views a week. And I am not stopping until we get there. Watch me.