_
What breaks when a Telegram founder community crosses 2,000 people_
I co-founded Startup4Nation two years ago. It is a 2,000-person community of early-stage founders and students in India. Most of the conversation happens on Telegram. Some of it happens on Luma. We run founder mixers, demo nights, and an annual pitch event.
This post is about the failure modes I hit between 1,000 and 2,000 members. Not the wins. The breaks. If you are scaling a community past 1k on Telegram specifically, these are the things nobody warned me about.

Topics stop being a routing mechanism. They become a graveyard.
Under 500 members, a few well-named topics carry the conversation. People skim, find their lane, post. Past 1k, this collapses. New members do not read the topic list. They post in the first chat they see. You start the day with six "introductions" in random topics, two pitch decks in #general, and one founder asking for a cofounder in #mumbai when they are in Bangalore.
What I tried:
Pinned a welcome message in every topic with a one-line "post X in topic Y" rule.
Asked mods to redirect posts. They got tired by week two.
Set slow_mode = 60 in #general. People just moved to other topics.
What worked:
Killed topics that did not earn their place every month. I run a 30-day audit: if a topic averaged under three posts per day, it got merged or deleted. Went from 14 topics to 6. Activity per topic went up. New members could actually see the conversation.
One and only one "drop anything" topic. Called it #lounge. It is the only topic that does not require a category. Everything that does not fit #funding, #hiring, #events, #build-in-public, #mumbai, #delhi, #blr goes there. This is the single biggest relief valve in the whole group.
If you take one thing from this post, take that: pick one topic and call it the junk drawer. Naming it #lounge is fine. #random is fine. Just have exactly one.

1. Telegram does not give you moderation granularity. You have to fake it. Telegram has admins, and it has per-chat permissions. That is it. There is no "remove on third warning," no "slow mode per user," no "ban by join date." At 2k members with three volunteer mods, you will burn out your mods in three weeks. What I tried: Built a Python bot with python-telegram-bot. Auto-removed intro posts that did not match a template. Mods hated it because false positives were high. Added new joiners to a "trial" group for 48 hours before letting them into the main group. People hated it. Drop-off was 40%. What worked: A simple /rules command with a one-line response. New joiners get an automated DM with the rules in plain text. Most of them read it. Most of the rest get a single, polite reminder from a mod and self-correct. A weekly mod sync on Sunday at 9pm. 20 minutes. Three mods, me, agenda in a pinned message. We review (a) who is breaking rules, (b) who is doing great and should be promoted to mod-in-training, (c) what content we are missing. Without this sync the mods drift. With it, they stay. A mod queue channel that only mods see. Any reported message goes there. Mods handle them async over the week. Crucially, we agreed on a 24-hour SLA. If a mod cannot handle a report in 24 hours, it goes to the next mod. The lesson: at this scale, your tooling is not the bot. Your tooling is the meeting.
2. Luma helps for events. It does not help for conversation. I moved all event RSVPs to Luma around the 1,200-member mark. It was a clear win. Event pages, ticket types, waitlists, post-event emails, all solved. But Luma is not a community platform. It does not replace Telegram. The 5% of people who RSVPed on Luma and never showed up to events were the same 5% who would have ghosted anyway. And the 20% who showed up to every event without ever RSVPing on Luma were the same 20% who were already most engaged in Telegram. What worked: treat Luma as a calendar, not a community. The Telegram group is the community. Luma is the schedule. Cross-link them on every event: Luma page has the Telegram link in the description. Telegram event announcement has the Luma link pinned. Never pretend one replaces the other.
3. The hardest problem is not growth. It is graduating. We have about 30 founders in the group who have raised pre-seed or are at revenue. They do not need a 2,000-person Telegram. They need a private WhatsApp with 8 other founders at their stage. They also do not want to leave the community publicly. What worked for me: I never asked them to leave. I started a parallel #founders-only topic that auto-adds anyone who has raised or hit ₹10L MRR. It is opt-in. It has its own rhythm. The main group stays open and welcoming. The founders topic stays tight and useful. The mistake I almost made: turning #founders-only into a paid tier. That would have killed the whole thing. The 30 founders would have left, and the 1,970 others would have felt like the community was leaving them behind. Community is not a funnel. It is a place.
4. Things I would do differently if I started over Add a lurker topic on day one. Half of any large Telegram group is reading, not posting. Give them a topic where they can react without composing a message. Emoji-only responses allowed. Write a one-page community handbook before 500 members. Not a long doc. One page. Rules, topics, mod contact, how to report. Update it quarterly. Promote two members to mod-in-training at every 500-member mark. Do not promote based on activity. Promote based on the people they help in DMs. That is the actual signal. Stop chasing "engagement metrics." Daily active members in a 2k-person group will sit at 8 to 12% on any given day. That is fine. It is not a funnel. Do not optimize it into the ground. What this has to do with DevRel I am applying for Founding DevRel at Failproof. The role is to own developer community end-to-end for an AI agent company. Slack, GitHub, Luma, events, the works. If you are hiring for this kind of role and reading this: this is the kind of community I have run. Not a Discord with three regulars. A 2,000-person community with mods, topics, events, an off-ramp for the high-engagement members, and a junk drawer for everyone else. I am not a community manager who has run one meetup. I am a community builder who has hit the failure modes above and shipped fixes for them. If you want to talk, my Telegram is in my profile. Or just open an issue on the github-triage-agent repo I shipped this week. I am triaging my own DM. ***Built and shipped github-triage-agent this week — github.com/shekhar854/github-triage-agent. Small LangGraph + Claude agent that triages GitHub issues. README documents where Claude Code breaks in production. Shekhar Thathera — co-founder, Startup4Nation. Community manager, Aayo App. B.Tech CSE (AI/Data Science), IIMT College of Engineering, Greater Noida. Applying for Founding DevRel.