There's a specific feeling I stopped having, and I only noticed when it was gone: the 9pm dread of I haven't done my Reddit replies today.

If you've tried marketing by replying to threads where people ask for what you build, you know the failure curve. Week one is exhilarating: every thread is a potential customer, every reply gets an upvote, you're sure this is the channel. Week two, finding good threads starts feeling like a second job. Week three, you're five tabs deep at 11pm with ChatGPT open in tab six, and some part of you has already decided to skip tomorrow.

The channel didn't stop working. The workflow was never sustainable. Almost everyone who quits this channel quits the logistics, not the strategy.

Here's the version I run now. It takes about fifteen minutes, it happens once a day with coffee, and, this is the part I care about, it has survived contact with months of me, a person with the discipline of a golden retriever.

The routine

7:50am: the email finds me. I don't "remember to check the tool." A morning email shows up with the night shift's log: what got found, what got scored, what got drafted, what's queued. The pull-vs-push distinction sounds small. It's the entire difference between a habit and a chore. Chores require remembering. My calendar remembers nothing before coffee.

7:52: read the night shift. Overnight, the system watched every channel, scored what it found (buying intent, fit against my actual product, author quality, freshness), and drafted the ones above the bar. Everything below the bar got skipped, not answered worse, so the list in front of me is short. On channels I still run approve-first, I go top to bottom saying yes or no. On channels I've flipped to autopilot, there's nothing to decide; I'm reading what already went out and what came back.

7:55: edit where I still want a hand on the wheel. Approve-first drafts get my eyes: sometimes two words, sometimes half the draft. That's the trust dial doing its job. When a channel's drafts stop needing my edits for a few weeks straight, that's the signal to hand over the keys, and the editing time goes to zero without the quality going with it, because the same judge I kept agreeing with goes on rejecting anything generic after I stop looking.

8:02: the sends pace themselves. Replies go out from my own logged-in browser across the morning, at a human rhythm: per-account caps, jittered gaps, no two replies alike. Not a blast at 8:02. A trickle that looks like what it is, a person answering people.

8:05: read the receipts. This is the part of the routine that got better instead of shorter. Which reply got an answer, which conversation went warm, which thread from last Tuesday just turned into a signup. Every reply has a receipt, so the fifteen minutes ends on the only question that matters: is this working, and where?

Why the design choices matter more than the tool

You can build a manual version of this with a spreadsheet, a saved-search bookmark folder, and a calendar block. If you do, steal the design constraints, because they're what make it survive:

Batch, don't stream. Checking mentions as they arrive is how a 15-minute task becomes ambient anxiety smeared across a whole day. One batch, one sitting. The threads from 2am are still fresh enough at 8am. Thread half-life is hours, not minutes, and reply quality beats reply speed every time I've measured it.

Pre-rank, never browse. The moment the workflow includes "scroll and see what's interesting," you've lost. Browsing expands to fill all available time and then takes the evening too. Whatever does the ranking (a rubric, a tool, a VA), the founder should only ever see a short ordered list.

Cap it. My queue engages a maximum number of threads per day, on purpose. An unbounded queue trains you to dread opening it. A capped queue is a finishable task, and finishable tasks are the only ones that happen daily for months.

Hold the bar at any autonomy. The bar was never "a human looked at it." The bar is "would a careful human have sent this." Score the thread first and skip below the bar. Reject any draft that reads like a template. Send from your own account at a pace a person would keep, with no two replies alike. Platforms ban spam volume and copy-paste sameness, not tools; hold that bar and it holds at every setting on the dial.

The compounding part

Here's what surprised me after a couple of months of the routine: the fifteen minutes get more valuable, not less. Every send feeds back: which communities produce replies that turn into signups, which phrasings convert, which authors respond. The ranking gets sharper about my buyers every week, and the channels I've handed over keep improving without me in the loop. Month three's fifteen minutes are worth multiples of month one's, with identical effort.

That's the quiet argument for making the loop small and daily instead of heroic and occasional: compounding only works on things you actually repeat.

The internet does not need you all day. It needs your judgment for fifteen minutes, and your voice around the clock. Go have your evenings back.