Scan top posts daily for concepts that already went viral (e.g. a tweet 'algorithm simulator'), note ones you could build into your product, then ship and demo your version once a couple of months have passed so audiences have forgotten.
When it applies: product with a visual/demoable feature, audience on X
Pick a category with proven revenue, copy the core, and pick one dimension to be clearly better at (Postiz added AI-agent support to a commodity scheduler). He chose X growth tools because Tweet Hunter had shown demand. Once demand is proven, the remaining risk is execution, which you control.
Being the 10th-100th AI profile picture generator or scheduler gets ignored by both algorithms and people. Take what worked and make it weird or different so it has novelty.
He used AI to help analyze and craft tweets for his agency's X marketing, noticed tools like Tweet Hunter hadn't integrated AI, and built that, treating existing paid competitors as validation.
He thought creator-brand matching was saturated back in 2015, then noticed a newer player (Passionfruit) doing well with ready-to-book sponsorship packages for creators; he suggests flipping it into a sponsorship page for brands.