Freek Borghgraef

Freek Borghgraef's recurring argument is that engagement is an architecture problem, not a content problem. Across his writing on Kahoot, Duolingo, and Khan Academy, the same mechanism keeps surfacing: a behavioral event made visible to the user, and often to others, at the exact moment it matters. On Kahoot's countdown timer he is explicit that it "doesn't improve learning outcomes in any direct measurable sense" but instead "convert[s] a passive classroom into an active behavioral event," because "the social pressure changes the cognitive state" and "the cognitive state changes what gets retained" . Khan Academy's Mastery Challenge gets the same treatment from a different angle: the product withholds the challenge until the system predicts skill decay, because "a challenge that's always available gets ignored" and the timing itself is "load-bearing" . The throughline is that when to intervene is as much a design decision as what the intervention contains.

A second theme is that streaks and points are not manipulation but identity-building, and he pushes back on the cynical reading directly: "That's not manipulation. That's the architecture of identity" . His point is that the number the app displays stops belonging to the app and starts belonging to the user, who then works to protect it, whether that's Duolingo's 84-day streak driving an 11 PM session or, in a much older example, BBVA's 2012 gamified banking app where "customers were completing financial health reviews they'd ignored for years, because there was a badge waiting on the other side" . He generalizes this into a portable mechanic: "Every app category has the same mechanism available. A ride streak. A savings streak. A workout streak. A reading streak" . The common move he identifies is making a behavioral count "countable, visible, and worth protecting" before it breaks, which is why he singles out Duolingo's Streak Freeze as "buyable before the behavior breaks, not after" .

On reward design specifically, Borghgraef's Khan Academy analysis lays out a differential-points argument he repeats across posts: video views earn less than exercises, and a mastery challenge "where the system tests a skill after a deliberate time gap, earns the most," which teaches learners what mastery requires "through the point value, not through an instruction" . He frames this as a corrective to a common failure mode: "Most engagement layers do the opposite. They give identical XP for passive consumption and active practice, then wonder why users complete content without retaining anyth[ing]" . The underlying design principle he keeps stating in different words is to "reward the behavior that predicts long-term retention, not the behavior that's easiest to complete" .

His fintech commentary extends the same logic from learning apps into banking and payments, and here his sharpest distinction is between competing on price and competing on identity. On Monzo he argues users didn't switch "because the FX rate was 0.4% better" but because the coral card "said 'I found something better' in a way a Visa debit card never could," concluding that "features get matched in a quarter. Identity takes a decade to copy" . He applies the same frame to Cash App's rotating daily Boosts, arguing they are not discount giveaways but a retention mechanism: "the behavior it buys is the swipe itself," and the real product being purchased is "a reason to open the app tomorrow" . He explicitly contrasts this with what he sees as the default, weaker fintech playbook of "more cashback" or "higher APY," calling both "price wars" that give the user no reason to return "before they've done anything" . The BBVA case functions as his proof that this identity-over-price logic predates fintech's current wave entirely, since a "150-year-old bank" got the same effect from badges and leaderboards decades before streak mechanics were common .

A more recent addition to his thinking is a concern about AI-generated app supply diluting the whole category. Anticipating the objection that "vibecoding" would erase demand for engagement tooling, he argues the opposite: "The problem becomes bigger. Exponentially" . His reasoning is that low-code and AI tools let non-specialists ship "a functioning app," but "functioning is not the same as fun & usable," so average app quality "PLUMMETS" while the app drawer gets more crowded, making relevance harder to achieve, not easier . He backs this with a stated shift in his own numbers, from "70% dead users after 3 months average" when he started to "80% and growing" now, and reports "3x as many publishers contacting us to improve app usage and user relevance compared to last year" . His conclusion is a direct imperative rather than a hedge: "UNDERSTAND WHY people use your app. Tailor the experience accordingly" .

Taken together, the concrete takeaways in his own words are consistent: make behavioral events visible rather than passive , time interventions to predicted user need rather than a marketing calendar , let users build a record or score they feel is theirs to protect , weight rewards toward the behavior you actually want repeated instead of the behavior that's easiest to log , and treat identity and relevance, not rate or feature parity, as the durable competitive layer in crowded categories .