Platform Β· Milestone Companion: The Flywheel Beats the Viral Hit
What this milestone asks
Publish one piece of content per week for four consecutive weeks β video or essay β and accumulate at least ten genuine interactions across them. The stage behind it: one primary platform plus one distribution channel; the content flywheel (real project β tutorial β Q&A/short-form β collected questions β next topic); frequency over virality; every answer you give becomes a FAQ entry feeding your topic pool. The bar is deliberately modest: the system that makes four weeks and ten interactions inevitable is the real goal.
Why this milestone
Stage 2 taught you to make one good artifact; Stage 3 teaches you to operate a practice, because influence compounds through cadence, not spikes. One piece per week builds more trust than one viral hit, for a simple reason: trust is a prediction. An audience that has watched you show up on schedule ten times predicts an eleventh, and that prediction is what makes them recommend you, invite you, and eventually β in Stage 4 β listen when you address enterprise audiences.
The ten-genuine-interactions bar is equally deliberate. "Genuine" excludes likes, courtesy comments from friends, and your own replies. It means questions, disagreements, corrections, retellings β signals that a stranger actually processed your work. Ten such interactions across four pieces is enough to run the flywheel: each becomes a FAQ entry, a fact-ledger correction, or the seed of next week's topic. An evangelist with no interactions is broadcasting; one with ten is in conversation, and conversation is where topics come from.
How to use the resources
Choose platforms by audience, not preference. One primary platform where your target readers already live, plus one distribution channel pointing back to it. For Chinese-language technical audiences, Bilibili (video) and Zhihu (long-form) are the established homes; for international reach, YouTube plus a blog or newsletter. InfoQ and QCon matter later β they are channels you graduate into, not where you start.
Study how a massive content operation is organized. The Berkeley RDI LLM Agents MOOC β led by Dawn Song, Spring 2025 concluded, videos still available β is worth studying as an operating model: a guest lineup that borrows credibility, assignments that generate participant output, certificates that create commitment, a Discord community that turns audience into participants. Note what you can steal at one-person scale: a fixed weekly slot, a community space, visible participant artifacts.
Build the flywheel before week one. The critical input is the real project β you must be building something, however small, or the well runs dry by week three. Set up capture infrastructure first: a running FAQ document where every question anyone asks you β comments, DMs, work β gets logged within a day. Papers with Code and the Hugging Face Discord/Forum are good listening posts for what practitioners are actually confused about; confusion is topic gold.
Plan the four weeks as one arc. A defensible sequence: week 1, a tutorial from your current real project; week 2, a Q&A piece answering the most common question from Stages 1β2; week 3, a "what I got wrong" piece β corrections earn disproportionate trust; week 4, a synthesis linking the previous three and announcing what comes next.
Keep the Stage 1 fact ledger open while you publish. Weekly cadence creates pressure to assert the unverified. Resist it: the Stanford HAI AI Index report remains your authoritative source for field-level numbers, and anything untraceable stays out or is explicitly labeled personal experience. Attend local meetups during this stage even before speaking at them β they teach you what your regional audience actually asks.
Common failure modes
- Two viral-hunting weeks followed by silence. Frequency beats virality because algorithm and audience both reward predictability. A piece engineered to explode usually explodes nothing.
- Platform sprawl. Publishing natively on five platforms means burning out by week three with five thin archives instead of one thick one.
- Counting vanity interactions. Likes and "nice!" replies are not in the ten. If you are unsure whether an interaction is genuine, it is not.
- Answering questions ephemerally. Replying helpfully in a comment thread and moving on, instead of promoting the answer into the FAQ. Every unpromoted answer is a topic you must invent from scratch later.
- Content without a project behind it. Tutorials about tutorials. The flywheel's first gear is a real project; without it the mechanism free-spins.
Acceptance checklist
- One primary platform and one distribution channel are named, with the reasoning written down.
- Four pieces (video or essay) are published, one per calendar week, four weeks in a row, with public timestamps proving cadence.
- At least ten genuine interactions (questions, substantive comments, corrections, retellings) are logged, each linked to its source.
- A FAQ document exists containing every question received, with an answer or a plan to answer.
- At least two of the four pieces trace back to a real project you are building.
- Every factual claim traces to the fact ledger or is labeled as personal experience.
- At least one piece's topic came from a collected question rather than your own initiative.
Going further
Four weeks is the proof; twelve weeks is the practice. Extend the streak before judging results β the compounding is back-loaded. Track interactions per hour of effort, and let that ratio, not raw views, steer topics. As your FAQ grows past twenty entries it becomes a course outline in disguise β Stage 4 material. Then approach local meetups as a speaker-in-waiting: you have a ten-minute recorded talk and a growing FAQ, both verifiable. That is the bridge from publishing to the stage.