why i moved installer access into a paid discord

why i moved installer access into a paid discord is a question i've answered a few times this week, so here's the full version.

short story: i had a working installer. i could have dropped a public download link, tweeted it, and watched a counter go up. instead i put access behind a private, hand-approved discord — a build room, not a community. if you want the short reason, it's this: a public download link hides the user, and i can't sharpen an agency ai workspace i can't see anyone using.

public downloads hide the user

a public installer feels like progress. it isn't, not at this stage.

think about what you actually get back from a download link. a number. maybe a region if your analytics are good. you don't see what they tried to do, where they stalled, which pane they opened first, what made them quit on day two. the download fires and the person disappears into the dark.

for a mature product that's fine. you already know the workflows, you've already closed a hundred feedback loops, the installer is just distribution. but i'm not there. aios is an ai coding agent workspace where the whole bet is that the work shouldn't live in fragments — one agent in a terminal, docs in a browser, notes somewhere else, the plan buried in a previous session. if that bet is wrong, or right in a way i didn't predict, a download counter will never tell me.

so a public link at this stage isn't distribution. it's me throwing the thing into a well and listening for a splash.

the paid discord is the feedback loop

i cleaned my discord down to almost nothing. not fifty dead channels pretending to be a community. one lobby. one build room. one workflow room. one product room. one setup room. that's it.

the structure does one job: a user shows their actual workflow, then aios helps them build faster. the room exists so i can watch the mess.

that's the part people miss. i don't want a forum where users help each other so i never have to show up. i want the opposite — i want to be in the room when someone says "my ai work is scattered across claude, codex, cursor, six browser tabs, two terminals, and a note i wrote yesterday." that sentence is worth more than a thousand downloads, because it's a spec. it tells me exactly what the agency ai workspace has to absorb.

paid (or hand-approved, which is its own kind of toll) does something free-public can't: it filters for people who'll actually wire ai into real work. a friction gate at the door isn't there to make money this early. it's there to make sure the people inside are builders, not lurkers. lurkers cost you nothing and teach you nothing. one builder showing a broken workflow moves the product.

why the first cohort matters more than traffic

i'm capping the first room at 100 founders, manually approved. people assume that's scarcity marketing. it isn't. it's bandwidth.

i can read 100 workflows. i can sit with 100 people and watch where the ai agent command center helps and where it gets in the way. i cannot do that with 10,000. if i open the floodgates now, every signal i actually need gets buried under support tickets from people who were never going to use the thing.

the first cohort is the design partner pool. they're not customers in the normal sense — they're the people whose friction becomes the next version. so i'd rather have 100 builders i can actually talk to than 5,000 anonymous installs i can't learn anything from.

raw traffic is a vanity metric until you've earned the right to want it. you earn that right by getting the product so sharp that a stranger can install it and succeed without you in the room. you don't get there by collecting strangers early. you get there by collecting the right ten, then the right hundred, and fixing everything they trip on.

and there's a second thing traffic can't buy: proof. posting more isn't the hack — posting proof is. if i say "ai agents can run your business," nobody cares. if i show the loop that wrote the article, deployed the site, cleaned the discord, and queued the content, now there's something to inspect. the first cohort sees that loop from the inside. they become the proof, not the audience for it.

how onboarding should feel

here's the standard i'm holding the room to: onboarding should feel like someone who already knows your context sat down next to you.

not a docs wall. not a 12-minute setup video. not "join our community and good luck." you come in, you show me how your work is actually scattered today, and the room helps you rebuild that as panes that stay open — an agent pane for the task, a terminal for commands and logs, a browser for the app, a file pane for the project, a scratch pane for decisions. the boring part is the product. the magic is that you stop paying the tax of rebuilding context every time you switch apps.

a public installer can't give you that. it can give you a binary and a readme. the readme doesn't know your repo, your stack, or the three tools you're already fighting with. a hand-approved room does, because i'm in it and i asked.

if you want to see the shape of the thing before you ask for a seat, the use cases page lays out the actual workspaces — claude code for deep messy reasoning, codex for fast repo-native execution, the surfaces around them. the demo shows the ai agent command center running. those exist so the room conversation can start from "here's my mess" instead of "what is this."

when to graduate to public installers

i'm not against public downloads. i'm against shipping them before the loop is closed.

here's my honest bar for opening the gate wider:

• a new person can install aios and hit their first real win without me in the room
• the top five workflows people bring are ones the product already expects
• onboarding survives my absence — the room teaches itself
• the support questions repeat, which means they're documentable instead of design problems

when those are true, the friction gate has done its job and it starts costing more than it returns. that's when a public installer stops being a well i'm throwing things into and becomes actual distribution. that's when traffic finally means something, because every install lands on a surface that's already been beaten into shape by the first hundred.

until then the gate stays. not to be exclusive — to stay close.

i won't promise you this is the only right way to launch. plenty of tools ship public day one and figure it out live. but i'd rather learn from a hundred builders i can name than a counter i can't interrogate. early distribution should buy you signal, not applause.

if you're wiring ai into your actual workflow — not prompt-dumping, actually operating across terminal, browser, files, and agents — that's the room. start by walking the use cases to see if the shape fits how you work, watch the demo, then ask for one of the first 100 seats in the build room. come show me the mess. i'll use it to make aios sharper.