You shipped something people can actually use. A few hundred or a few thousand of them showed up. What you still lack is an operating system for growth: a way to walk the product, capture every idea before it dies in chat, read the funnel, and ship only the loops that move a real action.
That is the job of AI bots for business growth in this method. Roshan had just launched Thursday Arena — an auto-battler you play with marketplace bots, X login, a leaderboard, and Elo. Vincent, who works on growth and is a Grokbot power user, treated every screen as a decision: what should the user do next, how fast can you ask, and which bot will log, measure, or chase the idea.
When you finish, you can stand up a growth bot that writes a Notion playbook, a metrics bot that pulses the product, an X workflow that turns follows and public feedback into usage, and a ranking rule so you do not spray users with every mechanic at once.
Key takeaways
- Spin up a named growth bot first. Talk through the product out loud. Have the bot log ideas into a Notion growth playbook while you walk the funnel.
- Growth work starts with the end-to-end experience, not a campaign calendar. If you cannot see login, practice, core loop, win state, and return home, you are guessing.
- Even with a lot of eyeballs, name the one action you want. Follows, practice rounds, and stream viewers are not the same as X sign-ins or return sessions.
- Product-led growth lives in lifecycle nudges inside the product. Sales-led growth is mass email, outbound, and LinkedIn. Pick the motion that matches the business you actually have.
- Voice into a bot is enough to start. Transcription will be messy. The agents can still file the idea if you keep talking.
- Naming a bot “data scientist” does not make it one. Tell it to go learn the craft, then keep those traits as a persistent skill.
- Rank by impact and effort, ship a top three, and stop. The failure mode is hitting users from every direction with things they did not ask for.
- Get a baseline of product-market fit before you scale acquisition. Marketing creates awareness. Growth maximizes the best parts of the product after people are in.
Who this is for — and who should skip it
This is for a founder, operator, marketer, or ops lead who already has a product surface — a site, an app, a game, an X account — and can walk a real user path. The stack assumed here is Grokbot (including an X connector), Notion, and a product that can show practice versus logged-in behavior. Slack shows up as a feedback sink. You do not need a full growth team. Vincent’s point is that growth is often zero-to-one work you do yourself until you prove it deserves more scope.
Skip this if you have no product to walk, if you are still hunting for an idea, or if you want a paid-ads or enterprise sales machine. Vincent is explicit that outbound is easy to get wrong, and that growing a game (or any product) before it is fun is the wrong order. If you cannot tolerate a human in the loop on DMs, email, and what actually ships, this method will push you to over-automate.
Capture growth ideas into a live playbook
The first job is not a brainstorm. It is a recording system. While Vincent watched the product, Roshan created a bot named Vincent Growth, told it to start logging growth ideas in Notion, and asked it to make a growth playbook. From that point, every useful sentence became an entry instead of a forgotten riff.
Prepare three things: a Grokbot you are willing to name after the function (growth, not a cute mascot), a Notion doc the bot can write to, and a way to talk — voice memo is enough. You do not need a perfect transcript. Roshan noted the capture was sloppy and full of “how about that,” and still trusted the agents to file it.
Walk the product with someone who thinks about funnels. Narrate what you see. When they flag an opportunity, say it again in one sentence and send it to the bot. Good output is a playbook that later shows up as a clean list with impact and effort scores — Roshan had the bot format the Notion doc that way — not a wall of raw chat.
Failure here is collecting ideas in the air. If the bot is not writing while you talk, you will remember the clever mechanic and lose the boring, high-leverage one (the share button on the win screen, the popup after three practice games).
Walk the funnel and design the loops
Vincent asked to see the end-to-end funnel, not a feature tour. That is the working rule: absorb the path a new person actually takes, then attach growth to specific moments.
Map the path before you add growth
Thursday Arena’s path was: land, log in with X (or skip into practice), authorize the app, see a leaderboard with stack rank and Elo, enter a shop with gold, set a lineup of bots with attack, health, and abilities, auto-battle over three rounds, restock between rounds, hit a win screen, return home. Your path will differ. The discipline does not: write down every screen and whether the user can brag, invite, convert, or get stuck.
They already had a useful split: practice with no account, and X sign-in for the real ladder. Roshan did not want a login wall just to try the game. Vincent treated that as a funnel, not a philosophy argument. If practice is unbounded, you may never get the identity you need for leaderboards, challenges, and DMs.
Share loops at the moments people already feel something
Vincent’s first instinct on the leaderboard was a share popup, arcade-style, as soon as you see your ranking: post this on X and brag. Same idea on the win screen — a share button when you actually won. He also wanted a share inside the match: because cards have scarcity tiers, sharing a bot or card could unlock a rarer version if someone else comes in. That ties rarity to the loop instead of bolting a generic “share this” onto the homepage.
Direct challenges were the social version of the same idea. Can Vincent specifically challenge Roshan? The other player should have to accept, Words with Friends-style. That opens “invite your friends.” Roshan liked it and also flagged it as possibly complicated. Log it; do not pretend it is as cheap as a share button.
Other loops they captured while still in the game: skip the shop or play rounds on a streak for more gold; let practice users win more often so the first sessions feel good; add difficulty modes for simulated opponents. None of these were proven. They were hypotheses attached to a screen.
What good looks like
A good pass is a short list of loops tied to moments: leaderboard share, win share, card share-for-rarity, challenge a specific player, practice-to-signup popup. A bad pass is a feature dump. Vincent’s later warning is the test: if you shipped all of it, you would be hitting users every which way with something they did not want.
Turn X attention into product usage
Thursday Arena had a site and an X account. Vincent’s view is that not enough teams treat those as real surfaces. The stream had concurrent viewers. That still is not growth if a first-time follower never gets a next step.
Speed to lead is so important. You told me you’re following me right now. That means you’re interested.
The concrete move: when someone follows @ThursdayArena, automatically DM them, thank them, send them to the game, and optionally give 10 free gold. Followers become a traffic source, not a vanity count. Roshan restated the model back: X is the platform; followers are how you drive traffic into the product.
Second move: listen in public. Vincent suggested using the X API to find people talking about Thursday Arena — thoughts, critiques, reviews. Grokbot already had an X connector (they called it the X MCP). The instruction they ran was to search feedback, summarize it in a new Notion section, keep user handles (needed for DMs), and focus on product feedback.
Acquisition was narrow: the stream and posts on X. The X workflow is how you widen that without pretending you have a full marketing engine. Good output is a Notion section of product feedback with handles, plus a welcome DM that asks for one action. Failure is broadcasting into the void, or DMing people with no offer and no memory of why they followed.
Instrument the funnel with a metrics bot
Roshan already had a bot whose job is dashboards. Internally, when they launch products, he spins up a metrics bot and tells it to give a pulse every 15 minutes: how many users, how the feature is doing, what bugs are showing up. He described it as a data intern handing you paper reports, except it runs in Grokbot. Vincent added the same pattern on Slack: people ask for a slice, the bot plugs into the dashboard, you get the cut on demand.
The snapshot they pulled at 11:25 a.m. Pacific: about 2,200 practice rounds, about 1,000 X sign-ins, a leaderboard on the order of a thousand users, and new signups already tapering after the launch spike. They had started measuring practice-to-X-login. Early game-engine stats showed a harsh win/loss picture — on the order of four, from the first hour — and a game-designer critique bot had already said the game was too hard for launch.
Vincent’s funnel recommendations sat on those numbers:
- After two or three practice games in a row, show a popup to sign in with X and continue.
- Let practice users win more favorably so they feel the game, then want the real mode.
- Slice win/loss for practice versus logged-in. If practice is brutal, the signup popup is asking people to join a punishment loop.
- Consider difficulty modes for the AI opponents. They did not have them yet.
Good looks like a live pulse you did not have to build a warehouse to get, plus one funnel (practice → X login) you can argue about with evidence. Failure is staring at a launch spike and calling it growth, or optimizing signup when you still have no reason to need an account.
Monetize later — but name the levers now
Roshan put the team in the familiar posture: get users at all costs, figure out money later. He still wanted a point of view. Vincent said he is not a gaming expert. What he offered were patterns from games he has seen, as recommendations, not results.
Match-three games stay free and use cooldowns so they can show ads. Resource games make you wait hours, then sell or ad-gate the skip. The useful question: what would players want more of? Gold, rarer cards, more matches, faster shop cycles, a challenge into a specific rival. You may have to add friction to create something worth paying for. Vincent called that a fine balance.
They already had a moderated feedback form that landed in Slack, and even a phone number on the site. Vincent’s idea: if you report a bug or send feedback, grant gold. The deeper point he and Roshan agreed on was player store of value — something people care about, invest in, and might pay to keep.
Do not invent a paywall in this step. Do write down the scarce things in the loop. If nothing is scarce, you have nothing to reward, share, or sell.
Rank the list, then ship only the top three
By the end of the walkthrough the playbook included, among other items: sharing wins, sharing a bot card for tier rewards, challenging a player, a weekly clipper leaderboard screenshot, an avatar showcase, welcome DMs on follow, practice-to-signup popups, emails from Potato (a teammate with an audience) so the sender name is a face people already know, and a mailing list aimed at gaming-developer types.
Vincent is a fan of welcome DMs. For X signups, he liked practice two or three times, then the popup — high impact, low effort, upper-right quadrant. He also liked the share ideas, especially leaderboard status, even more if the game is hard: surviving it is the brag. Roshan treated those as the working top three and planned to feed them into their engineering path: throw agents at the ideas, take PRs, playtest.
Outbound got a harder look. Vincent’s view: most people are on the receiving end of outbound they did not want. The job is to get the thing read. He looks at the sender name, not the subject line and not a “you must open this” hook. So use names people have already seen on X, LinkedIn, or videos. Potato sending as Potato. They also talked about giving a bot its own email address and letting it run campaigns — something Vincent said he is experimenting with. For a colder list, he sketched reaching developer-ecosystem people at large gaming studios on LinkedIn, enriching work emails, and writing as humans building in public, not as a blast.
What they refused to do is equally part of the method. Vincent said the danger is adding too many things. Roshan agreed. Missing ideas were less important than restraint.
The danger is adding too many things for sure. We’re hitting users every which way with some random thing that they don’t want.
How to run Grokbot when growth is a one-person job
After the game, they talked about the job itself. Growth often means doing data work, UI judgment, and email empathy with no extra headcount, until you prove the motion. Bots are how Vincent covers those seats without pretending one prompt is a department.
A bot named data scientist is not automatically a good data scientist. The night before, he told a bot to take 101, 201, and 301 courses on being a strong data scientist. It found classes, pulled traits he cared about, and kept them as a persistent skill. He did the same to a designer after ugly output: go take courses on what makes a better designer. Roshan boiled the prompt down: go teach yourself how to do X, then be an expert on the thing.
Daily rhythm: a 9:00 a.m. pulse on what everybody is working on; then one-to-one chats with specific bots; then an end-of-day roundup that reloads context — which bots own what. Hybrid, not religion: one bot that sees the whole board, plus specialists.
Go teach yourself how to do X and then be an expert on the thing.
Vincent’s last growth principle was the one to tape above the playbook. Do not confuse marketing with growth. Marketing is top-of-funnel awareness. Once people are in, growth is maximizing the best parts of the product. And get product-market fit for the game (or whatever you shipped) before you obsess over scaling it. There is a chicken-and-egg, but you can still break the work apart.
Growth is also just how would I want to be treated as a user?
Roshan had joked that giving things away is a growth hack. Vincent did not turn that into a coupon strategy. He put it on the same axis as hearts and minds: if you are the user, what would you want, and how fast would you want someone to answer you?
Prompts, bot instructions, and workflows
These are cleaned from what they actually told the bots. They are not a different method.
Growth scribe, used while walking the product:
Start logging growth ideas in Notion. Make a growth playbook I can add to. Whenever I (or a guest) call out an opportunity, capture it as a playbook entry with enough context to ship later.X listening into the same doc (reconstructed from the speaker’s description):
Do a quick search of all feedback about Thursday Arena on X. Summarize it in a new section in your Notion doc. Include user handles. Focus on product feedback.Metrics intern (reconstructed from the speaker’s description):
Your job is, every 15 minutes, to give me the pulse. How many users are using it? How is the feature going? Are there bugs we are hearing?Skill-load a specialist (reconstructed from the speaker’s description):
Go take 101, 201, and 301 courses on being an excellent data scientist. Download the key traits of a good data scientist and keep them as a persistent skill in this bot.This output looks weak. Go take a ton of courses on what makes a better designer, then apply that bar to the next pass.Daily hybrid ops (reconstructed from the speaker’s description):
Every day at 9:00 a.m., give a quick pulse check: what is everybody working on?
At the end of the day, round up and refresh context: which bots are working on what.If you have three or more bot jobs, keep owners explicit so the 9:00 a.m. pulse has something to report.
| Use case | Bot / tool | Input needed | Output | Owner |
|---|---|---|---|---|
| Idea capture while walking the product | Named growth bot + voice + Notion | Spoken opportunities, screen by screen | Growth playbook, later scored for impact/effort | Founder or growth |
| Public product feedback | Grokbot X connector (X MCP) | Product name, focus on critiques/reviews | Notion section with summaries and handles | Growth |
| Live pulse | Metrics / dashboard bot | Usage, feature health, incoming bugs | Report every 15 minutes; on-demand slices via Slack | PM |
| Design / difficulty critique | Game designer critique bot | Current game rules and feel | Critique (example: too hard for launch) | Design |
| Specialist skill | Bot named for the craft, then trained | “Take 101/201/301 courses on X” | Persistent skill, better later output | Whoever is doing zero-to-one growth |
| Daily coordination | Hybrid: one pulse bot + specialist bots | What each bot is doing | 9:00 a.m. pulse, end-of-day roundup | Operator |
Measurement and business impact
No one in this session claimed a lift from the ideas. They claimed a way to see. The checkpoints they actually used:
- Volume: practice rounds, X sign-ins, leaderboard size.
- Funnel: practice → X login, including whether a popup after two or three practices should exist.
- Quality of session: win/loss, and whether that ratio differs for practice versus logged-in users.
- Difficulty as a growth input: a critique bot saying the game is too hard is a funnel issue, not only a design issue.
- Attention versus action: concurrent viewers and follows do not count until you specify the next action (open the game, accept a DM, share a rank).
- After the spike: they already saw new signups taper. That is when a playbook matters more than a launch thread.
Qualitative bar from Vincent: are you measuring marketing (did people hear about it) or growth (did people who are in get more of the best part of the product)? If you cannot separate those, your dashboard is an awareness report.
Pitfalls and guardrails
- Shipping the whole playbook. The stated failure mode is too many interrupts. Rank, take three, let agents PR and playtest those.
- Growing a product that is not yet good. Vincent’s first final thought was product-market fit for the game. A share button on a miserable loop advertises the misery.
- Login walls with no reason. They let people practice on purpose. Convert when you need identity, not because signups look like progress. Roshan noted they did not yet have a strong need to force sign-in.
- Practice that teaches people they will lose. If early win/loss is brutal, fix difficulty or bias practice wins before you nag for X auth.
- Outbound that reads like outbound. People skim sender names and drown in mail they should unsubscribe from. Use recognizable names. Do not lead with a frantic subject line.
- Monetization that is only friction. Cooldowns and waits can create a lever. They can also wreck feel. Name what players want more of before you gate it.
- Eyeballs without a targeted action. A live audience, a follow, a practice round, and a logged-in match are different events. Pick one per surface.
- Untrained specialist bots. Role names are labels. Skills have to be loaded and kept.
- Voice capture without a human pass. Transcription is imperfect. The bot can file the idea; a person still decides what users will see, what a DM says, and what an email from Potato should sound like.
- Challenges and invite-your-friends. High upside, higher product complexity. Do not treat them as the same effort as a win-screen share button.
A human stays in the loop on: what the welcome DM actually says, whether gold rewards for feedback are farmable, which three items engineering should take, and whether the game is fun enough to deserve more traffic.
7-day implementation plan
Day 1
Create one growth bot. Point it at a Notion playbook. Walk your live funnel the way Vincent did: land, auth or practice, core loop, win or fail, return. Speak every idea in one sentence. Do not prioritize yet. If you have an X account, write the single action you want a new follower to take.
Days 2–3
Stand up the metrics bot on a 15-minute pulse: users, the one funnel that matters, bugs. Add a practice-versus-logged-in slice if you have both modes. Run the X search into the same Notion doc, with handles, product feedback only. Read win/loss or the equivalent quality metric. If a critique bot (or a sharp user) says the product is too hard or too empty, log that as a growth blocker, not a later polish ticket.
Days 4–5
Score the playbook for impact and effort. Pick three. The set they favored is a useful default if it maps to your product: a welcome DM on follow, a share at the emotional peak (leaderboard or win), and a low-friction convert after a few successful practices. Send those to whatever ships code in your world — in theirs, agents that return PRs and playtests. Do not add challenges, clipper leaderboards, avatar showcases, and cold LinkedIn in the same batch.
Days 6–7
Load skills into the bots you will actually reuse (data, design, growth scribe). Turn on the 9:00 a.m. pulse and the end-of-day roundup. Draft, do not blast, any email that depends on a recognizable sender. Re-read the playbook and cut anything that would poke the user without a reason. Check the original question: are you buying awareness, or making the best part of the product happen more often for people already in?
Start with the playbook, not the full board
The business outcome is not a longer idea list. It is a product people can find, finish, brag about, and return to, with bots doing the filing, the pulse, and the first draft of the experiments — while you protect the user from your own enthusiasm. Watch the walkthrough to see the funnel, the voice capture, and the live ranking, then start with one growth bot and document that first workflow in Notion before anything hits production.
FAQ
Do I need Grokbot, or will any AI bot work?
The method was run in Grokbot: named bots, voice into a playbook, Notion writing, an X connector they called the X MCP, recurring pulses, and agents that take ideas toward PRs and playtests. Other tools were not compared. If your stack cannot write the doc, search X, and run a 15-minute pulse, you will be doing those jobs by hand.
Product-led growth vs sales-led outbound: which one is this?
Vincent’s split: it depends on the company. Product-led work is lifecycle — you saw someone struggle, so you put a nudge there. Sales-led work is mass email, outbound, and LinkedIn. Thursday Arena was treated as PLG with X as the distribution surface. Outbound was optional, narrow, and easy to get wrong.
What is the difference between marketing and growth here?
Marketing is top-of-funnel awareness. Growth starts after people are in, when you maximize the best parts of the product. Vincent wanted that distinction explicit so teams stop asking “how do we grow it immediately” before the product is worth returning to.
Should I force signup or let people practice first?
They let people practice without an account on purpose. Vincent did not reverse that. He added a convert: after two or three practices, popup to sign in with X, and make practice winnable enough that people want the real mode. If you do not yet need identity, do not pretend a login wall is a growth strategy.
Is outbound worth doing?
Only if people will actually read it. Vincent’s rule of thumb is that recipients look at the sender name, not the subject line. Use a face they already know, or do not bother. Cold lists of gaming-studio developers were a maybe for this specific product, not a default motion.
One general bot vs many specialist bots?
Vincent uses both. A morning pulse from a generalist, then one-to-one work with specialists, then an end-of-day roundup so context does not fragment. Role names are not skills. Teach the specialist, then keep the skill.
When should I think about monetization?
You can chase users first, but Vincent still wanted the team to name levers players would want more of. In games that often means cooldowns, waits, rarity, or gold — which means some friction. Create store of value before you add a paywall. Reward feedback with gold was one small example, not a business model.
What should I ship first?
High impact, low effort, tied to an action you already need. Their working three: welcome DMs for new follows, share at leaderboard or win, and a practice-to-X-signup popup. Everything else stays in the playbook until those exist and the product is not too hard to recommend.