Communities worth being in
Discord, X, hackathons, local meetups. Where the actual hiring conversations happen and where the field's signal lives — and where it doesn't.
Prerequisites
Stack
X account, Discord, GitHub, lu.ma or partifula local meetup app or your school's CS group
By the end of this module
- Pick the two communities worth being in this year — and refuse the other ten.
- Use X without becoming someone who lives on X.
- Find a Discord that genuinely accelerates your work.
- Win or place at a hackathon by picking the kind that actually has good odds.
Almost every offer, intro, and opportunity that mattered in my career came through the same channel: a community I was already in for non-career reasons, where someone happened to know I was looking. None of them came from a job board. None came from cold-applying. The exceptions were the offers that turned out to be bad.
This is not a “build your network” lecture, which is the most useless career advice ever written. It’s the specific claim that in 2026, four kinds of community matter for engineers — and being meaningfully present in two of them, for two years, replaces almost everything else you might otherwise spend that time on. Not “active.” Not “follower count.” Present. Recognisable. Helpful.
Most CS students get this wrong in two opposite directions. They either ghost every community (“I’m too busy to be on Discord”) or they’re nominally in fifteen and meaningfully in zero. The fix is on this page.
Set up — the four-archetype audit
Before you read further, write down every online community and group you’re currently a member of. Discord servers. Slacks. Subreddits. X follows. School clubs. Mailing lists. Be thorough — most students underestimate this number badly. Now mark each one in one of three buckets:
- A — I’m meaningfully present here. People know me. I learn things. I help people.
- B — I lurk and learn but don’t participate.
- C — I’m in the channel but it’s noise. I don’t even read it.
Most students have 0–1 in bucket A, 3-5 in B, and 10+ in C. Almost everything in C should be muted, archived, or left. The audit takes 20 minutes and changes how the rest of this module reads.
Read these first
Three sources, in this order, then stop:
- Patrick McKenzie — Networking Doesn’t Have to be Sleazy. post · 20 min · the cleanest framing of why “networking” is the wrong word for what actually works.
- Cate Huston — On building a public profile. post · 10 min · honest about the costs and the asymmetric benefits.
- Visakan Veerasamy — On befriending strangers on the internet. post / thread · 15 min · a master class in being a good citizen of an online community, by someone who is one.
Stop. Skip the “LinkedIn networking” books. They are written for a 1995 networking model that has not survived the internet.
The four community archetypes
Each does a different job. You should be in two — one from the first pair, one from the second.
| # | Archetype | What it’s for | How to use it |
|---|---|---|---|
| 1 | X / Twitter | Following the field’s signal in real time | Follow 50 great accounts; mute 200 noise ones |
| 2 | Discord servers | Hands-on help, project-buddies, real conversation | Pick 1-2; show up weekly |
| 3 | Hackathons | Project-completion practice, team-finding, recruiter contact | Schedule 2-4 per year |
| 4 | Local meetups + school orgs | Warm, in-person intros that cold internet can’t replicate | One regular thing, in person |
Pair one signal-following community (1 or 4 — passive intake) with one active-participation community (2 or 3 — visible work). That’s the minimum. Adding more before you’ve gotten value from these two is just churn.
X for engineers — a strategy that doesn’t ruin your life
X is the most signal-rich and the most attention-destroying place on the internet. Most engineers either use it badly (doomscrolling, hot-take consumption) or refuse it entirely. The middle path is real:
- Follow 50 people whose work you’d want to do, mute 200 noise accounts. The follow side is the obvious half. The mute side is the half that makes it tolerable. Mute aggressively: politics, drama, anyone who’s mostly retweeting, anyone whose job seems to be having opinions.
- Read once a day, max twice. Set a time. 20 minutes. Then close the app. The “I’ll just check…” pattern is what destroys the value of X for almost everyone.
- Post once a week, max. A real post — a small thing you built, a screenshot of a thing that worked, a question you actually have. Not a take.
- Reply more than you post. A thoughtful reply on a small account’s post is one of the highest-leverage internet actions there is. The poster will remember you.
- Don’t engage with viral threads. Your reply will be invisible, the discussion will be bad, and you’ll lose 30 minutes you can’t get back.
The starting follow list for an ML/eng student in 2026 looks roughly like: Karpathy, Simon Willison, Charity Majors, Lilian Weng, Andrej Karpathy, Eugene Yan, Julia Evans, Mitchell Hashimoto, Steph Smith, Patrick McKenzie, Dan Luu, Tanya Reilly, Will Larson. Plus 30-40 others you’ll find naturally as you read.
If X actively makes your week worse — which it does for some people, no shame in it — switch this slot to a great RSS reader and high-signal newsletters (Stratechery, Last Week in AI, Import AI, The Pragmatic Engineer). You don’t need X. You need the signal that flows through it.
Discords worth joining
Pick at most two. Be present in both. Five worth considering:
| Discord | What it’s actually good for |
|---|---|
| Hugging Face | The hub for ML engineers and researchers. Ask real questions, get real answers. |
| fast.ai (forum, not Discord, same idea) | The most welcoming ML community there is. Beginner-friendly without being shallow. |
| Anthropic / OpenAI builder communities | Where people actually building with LLMs hang out. High signal, low ego. |
| Language-specific (Rust, Go, Python, TypeScript) | Pick whichever language you live in. The Rust Discord especially is a model of healthy community. |
| Your favourite OSS project’s Discord | The one you picked in 06.2. Hang out where the maintainers are. |
The pattern that works in all of them: lurk for a week, watch how people ask and answer, then start participating. Help someone whose question you can answer — even if your answer is just “here’s where to look.” Most people in Discords are too shy to answer questions; you’ll be remembered fast.
The pattern that fails: joining 12 Discords, never opening any of them, occasionally posting a low-effort question and getting no response.
Hackathons — pick the right kind, not the biggest
Hackathons get talked about as a single category. They’re not. The category you pick changes everything.
| Type | Effort | Win odds | Recruiter density | Worth it for |
|---|---|---|---|---|
| Local school hackathon | Low | High (50-100 teams) | Medium | First-time builders, finding a team |
| Major university (HackMIT, TreeHacks, HackTX) | High | Medium | High | Prestige line on resume |
| Themed / niche (LLM hackathon, ETHGlobal) | Medium | High (smaller field, narrower judging) | Very high | Specialists; the highest ROI |
| MLH generic | Low | Low (huge field, vibe judging) | Medium | Practice; not strategic |
| Company-run (Anthropic, AWS, etc) | Medium | Medium | Maximum (recruiter is the judge) | The single most underrated kind |
The big take: company-run hackathons are dramatically underrated. A small, themed, sponsored hackathon with 30 teams and Anthropic engineers as judges is worth 10 generic MLH events. The recruiters are literally the judges. You’re being interviewed for free.
What wins, broadly: a working demo on a narrow problem, a clear story you can pitch in 90 seconds, and one piece of polish that makes the project memorable (a real domain, a slick UI, a surprising data finding). Not: technical complexity for its own sake, half-finished ambition, or anything that depends on the wifi working during demo time.
Local meetups + school orgs — the warm-intro engine
The internet has trained us to forget that an in-person conversation is worth twenty Discord exchanges. Find one regular in-person thing in your city or campus. Show up monthly for a year.
For students:
- Your school’s CS / AI / hackathon club. Even if it’s not impressive, be one of the regulars. The other regulars are your warm-intro pool for the next decade.
- A
lu.maorpartifulfor AI / ML / eng meetups in your city. Most cities have at least one good monthly one in 2026. - If your city is tiny: start one. Three people in a coffee shop counts. The act of organising is itself the credential.
The question to ask, every month: “Whose work am I most curious about?” Then introduce yourself to that person, briefly, and ask one specific question about it. That’s the entire “networking” skill.
You can ignore LinkedIn
Yes, really. LinkedIn is a job board and a place where recruiters DM you; it is not a community. Have a clean profile, accept legitimate connection requests, ignore the “thoughts on” content, and don’t post there. The opportunity cost of any time you spend posting on LinkedIn is time you could spend on a Discord helping someone, which compounds 10x more.
The one exception: if you’re applying to a specific company, look up the engineers there on LinkedIn before the interview. Two minutes per person. That’s its real use.
Five actions this week
- Audit your existing communities. Mark each as A/B/C. Leave or mute everything in C.
- Pick one signal-following community. Most students should choose X with the curation rules above. If X is bad for your mental health, pick high-signal RSS / newsletters instead.
- Pick one active-participation community. Either a Discord (one of the five above, joined this week) or a regular local meetup (committed to attend the next one).
- Schedule the next 6 months of hackathons. Pick 2-4. Put them on your calendar now. Block the weekends.
- Help one person publicly. Today. Pick one question in your chosen Discord and answer it well. The one-rep training day for the rest of the practice.
Repeat this audit every quarter. Most students drift back into 12 channels of noise within 6 months; the regular audit prevents it.
Going deeper
When you have specific questions, in this order:
- Anu Atluru — On finding your people. post · 20 min · the longer-form version of why community membership is upstream of career outcomes.
- Tanya Reilly — The Manager’s Path through Glue Work. post · 15 min · how to be the person communities remember without burning yourself out.
- Indie Hackers communities — for a flavour of what active, friendly online communities look like when they work.
- Cal Newport — Deep Work (chapter on the principle of least resistance). book · skim · why most communities default to noise and what to do about it.
Skip “growth hacking” content for personal brand. None of it scales below the celebrity-engineer tier, and it ruins the people who try.
Checkpoints
If any one wobbles, the corresponding section above is what to reread.
- Name the four community archetypes. Which two are you currently in meaningfully, and which would you add?
- Walk through the X strategy for engineers. What’s the rule about replies vs posts?
- List the five hackathon types and their win-odds / recruiter-density profiles. Which two are the highest-leverage for your situation?
- Why is LinkedIn a job board and not a community? When is it actually useful?
- What are the five actions this week? Which one have you not done yet?
If you can answer all five, you’ve earned 06.4. Next stop: 07.1 — internships, the on-ramp to actually getting paid for this.