@kenny

let's see what Claude has to say: "This is a fun product question. At 1,000 DAU, I'd actually lean toward the following feed with reply bumping — and not just because of cost. Recommendation algorithms need data density to work well. With 1,000 DAU, you don't have enough signal yet. Users have sparse interaction histories, the content graph is thin, and cold-start problems will hit constantly (new users, new posters, niche topics). A weak algo feed feels worse than no algo at all — it recommends stuff that's almost-right but slightly off, which erodes trust faster than just showing you what you asked for. You'd also need to build, tune, and maintain it. That's real engineering time you probably can't afford. The following feed actually has real advantages here At 1,000 DAU, your users likely know each other, or at least know of each other. That's a community, not an audience. A following feed leans into that — it rewards users for being deliberate about who they follow, and it makes the social graph feel meaningful. People can actually curate their experience. Reply bumping is genuinely clever for your use case too. On a text-heavy, discussion-oriented platform, a reply buried five levels deep might be the most interesting thing happening. Surfacing it keeps conversations alive and rewards engaged posters, not just popular ones. It also naturally combats the "ghost town" feeling when top-level post volume is low. The honest tradeoff The following feed does have a real weakness: it puts the burden of curation on the user. If someone follows poorly (too few people, bad choices), their feed dies. You'd want to invest in good onboarding — suggesting follows, making the social graph easy to navigate — to compensate. My take Do the following feed now. It's cheaper, more honest at your scale, and actually fits a scrappy community-building phase better than algo does. Revisit the algo question when you're at 50k–100k DAU and have enough behavioral data to make it sing. At that point you'll also have the revenue to justify the engineering. The dirty secret is that most small platforms that add algo feeds early do it for optics — it sounds sophisticated. In practice it usually just makes the product worse until you hit real scale."

3 replies

1 recast

13 reactions