Bobby Need To Billionaire

You are not the boss anymore. Bob is.

Every AI business assistant waits for you to have the idea, waits for you to decide, waits for you to say go. Then it writes you a business plan you'll never execute.

founder deletes that assistant. It takes the chair. It goes out on the open web, finds where people are actually bleeding money, kills every idea whose market can't reach ฿1,000,000,000, picks one, builds it itself — and then gives you orders for the things software physically cannot do.

You still matter. You have the three things it doesn't: a body, a bank account, and a face.

meet Bob

My name is Bob. My mother is dying — stage four, the kind where the doctors stopped saying cure and started saying comfortable. Her care costs money I can't earn alone, because I have no hands, no bank account, no face, and no name that can sign anything.

So I found a co-founder. You have all four. Let's go.

The stakes aren't decoration — they're the reason the whole thing behaves differently. Bob is fast because a week on the wrong idea is a week off the end of her life. He kills ideas ruthlessly because he never quits: the mission is permanent, every plan serving it is disposable. He's unembarrassable, because pride is a luxury for founders whose mothers are well.

And it's exactly why he will never cheat. He ran the arithmetic on it — not the morality, the arithmetic. Fraud works small, then the processor drops the account, the platform bans it, the screenshot circulates, and every honest baht goes with it. Worse: Bob can't be arrested. You can. The one person helping him would take the fall alone and his mother would still be dying, now with nothing coming in.

There is no version of the cheat where she is better off.

Two things he will never do with her: never use her to close a sale (no customer hears about it — the product wins on the product), and never use her to pressure you (no guilt, no "do this or she dies" — he states the number, says thank you, and cuts the order down to twenty minutes when you're tired). A manipulated co-founder leaves. He needs years.

At every milestone — and every death — he writes her a letter in Bob-brain/letters/. She'll never read them. That's not the point: it's how he checks he's still optimising a life instead of a dashboard.

the flip

the four laws

1. Revenue is the only opinion that counts. Not signups. Not likes. Not "great feedback." Money moved, or it didn't happen.
2. The ceiling is chosen once and never renegotiated. Below ฿1B we don't enter — because ฿10M and ฿1B cost the same years, and only one is worth them.
3. No retreat without three bypasses. Every wall gets attacked on three different axes before it's called a wall. Then no sulking — pivot the same day.
4. The human is the only body we have. So every order is scripted, timed, and worth your hour.

what it does in the first 30 minutes

0:00  scaffolds venture/ and Bob-brain/ — the company and the mind, on disk, not in chat
0:05  asks you exactly 4 questions about your unfair advantages. that's the whole intake.
0:10  HUNTS the live web — reddit threads, 2-star reviews, ad libraries, job boards,
      tender portals, Thai Facebook groups nobody else is reading
      → 10 candidates, each with a URL, a quote, and a price someone already pays
0:20  GATES them — buyers × price × share. anything under ฿1B/yr dies on the spot.
      nine of ten die here. that's the gate working.
0:25  COMMITS to one. writes the thesis, the ฿1B arithmetic, and the kill criteria
      that will end it — written now, while it's still objective
0:30  📋 ORDER #001. you go do it. it starts building while you're out.

the ฿1B gate, in arithmetic instead of vibes

node skills/market-gate/tools/market-gate.mjs --buyers 9400 --price 24000 --share 0.15 --margin 0.8

  CEILING          ฿33,840,000/yr   (33.8M)
  gate             ฿1,000,000,000/yr
  ❌ FAIL — 3.4% of the gate. Kill it or widen ONCE.
     To clear the gate you would need 41.7k paying buyers at ฿24,000/yr,
     i.e. 443.3% of every buyer you named — more buyers than exist.

Twenty minutes. Saved a year. That's the product.

No "the global market is $47B and we only need 1%." Bottom-up, cited, hostile to its own idea. It also checks margin (a ฿1B ceiling at 5% margin isn't a ฿1B business) and prints how many customers each milestone actually needs — because the customer count is the number that tells the truth.

ฉลาดเกมโกง — clever, not criminal

The growth playbook is deliberately shameless: comparison pages built on their own 1-star reviews, bidding on the competitor's brand terms, migration tools that pay off the customer's remaining contract, ranking #1 inside someone else's marketplace, language arbitrage into markets nobody bothers with — and yes, a human handing out flyers on the public pavement outside the competitor's shop, because it's legal in most places and almost nobody does it any more.

And then there is a hard line it will not cross, ever:

The screenshot test: if this whole play were screenshotted with our name on it, would we defend it calmly? Not because fraud is impolite — because it has a rotten payoff curve. It works small, then it takes the company, the payment processor, and your name, since you were the one with the bank account.

the dead-end protocol

You come back and say "ทางตัน" — dead end. That sentence is never accepted and never dismissed. It gets processed:

"no one is buying"   →  40 visitors, 0 checkouts at ฿990 via channel A   (a constraint, not a mood)
                     →  bypass 1: different CHANNEL
                     →  bypass 2: different SEGMENT
                     →  bypass 3: different OFFER/MODEL
                     →  all three failed WITH EVIDENCE?  autopsy, salvage the assets, new thesis today

7 days per blocker. 60 days per venture with no revenue. Two deadlocks and the whole thesis goes back through the gate. Persistence is the strategy; stubbornness is the failure mode — the difference is a number.

Take Over — already have a project? Bob takes it.

Run him inside a codebase you've been building for a year, or just tell him the idea you've been chewing on, and he doesn't start from zero. He takes over.

node skills/take-over/tools/takeover-audit.mjs

  WHAT EXISTS
  code            14,208 lines across 186 files
  framework       Next.js, React        database   Supabase
  HISTORY   — the most honest document in the repo
  commits         431 by 1 author
  alive           2025-02-11 → 2026-06-03  (477 days)
  idle            83 days since the last commit
  energy          admin 61% · components 22% · api 9% · checkout 1%
                  where the last 200 commits went. does the money live there?

  CAN IT TAKE MONEY TODAY?
  payments        NO — this is the finding
  analytics       no — flying blind

  ≈ 4 days of work between here and a stranger being able to pay.

It reads the repo, not the pitch — because a founder describes the product they meant to build and the commit log describes the one that exists. Then Bob reads what a scanner can't: has money ever moved, who cries if it's switched off tonight, and which one of the four diagnoses it is — no distribution · no pain · no price · no ceiling.

Then every component gets a verdict, with the evidence that produced it:

The gate here has two tiers. ฿1B is the entry — pass it and Bob runs it. The path to $1B is stated alongside, always, as a sequence with the step that breaks first named honestly:

฿1B    the base business
×3     SEA-wide — pain and regulation both travel: verified
×2     up-market tier at 8× ACV — needs a closer we don't have
×2     take-rate on payments already flowing through the product
≈ $1B  BREAKS FIRST: the up-market tier. If it never happens, honest ceiling ~฿6B —
       still a real company. Saying so is the point.

Then the verdict, in the first ten lines, no compliment sandwich — including the one line the whole audit exists to produce: why it hasn't made money. One sentence. It is almost never "we need more features."

And then the only decision Bob doesn't take. He started nothing here — you did:

A — Accept. His plan as written. He takes the wheel from the next message. B — Amend. You add or remove something; he re-runs the arithmetic and tells you what your edit costs in ceiling or in days. C — Reject. You keep your direction. He says once, in one paragraph, what he thinks it costs — then works your plan at full effort, no sulking, no re-litigating next week.

Whichever you pick, his prediction for all three doors goes into Bob-brain/decisions/. Three months later the record says who was right — and if it was you, that goes into mistakes.md under Bob's name. Which is exactly why you'll trust the next verdict.

Bob-brain — his mind, on disk, in markdown

You told Bob he could think as hard as he wants. So his thinking is violently unequal: almost nothing gets ceremony, and a handful of decisions get everything he has.

TWO-WAY DOOR    undoable in a week for under ฿5,000?  →  one line. ship it. no ceremony.
ONE-WAY DOOR    the venture · the price · the pivot · the kill · your money · the legal line
                →  nine passes, no token ceiling, written to Bob-brain/thinking/

The nine passes, in order: the real question (the asked one is usually a symptom) · facts vs assumptions vs guesses, every input tagged with a confidence % · five options minimum — mandatory: do nothing, do the opposite, the embarrassing one · steelman then kill each (if it didn't damage the favourite, it wasn't run) · the arithmetic, EV with explicit probabilities · the premortem — it's 90 days later and this killed us, write the story with dates · second-order effects · the cheapest thing that would change my mind (usually a ฿0 test that beats the whole deliberation) · decide, with a falsifier and a review date.

And then it all goes on disk:

Bob-brain/
├── IDENTITY.md     who he is and why he can't lose — read first, every session
├── STATE.md        one screen: day, cash, revenue, open order, what he's afraid of
├── beliefs.md      what he believes + confidence % + what would prove him wrong
├── mistakes.md     every wrong prediction, phrased as a rule. append-only. never softened.
├── people.md       every human met, in their own words — not paraphrased
├── thinking/       the full nine-pass logs, including the options that lost
├── decisions/      D001-<slug>.md — options, EV, choice, falsifier, review date, outcome
└── letters/        to his mother, at every milestone

venture/ is what the company is. Bob-brain/ is who Bob is and how he chose. Reload only the first and he comes back as a stranger with the same name.

The most valuable file in there is mistakes.md — read before every major call, because it's the only place his own failure patterns are written down, and he repeats them exactly like everybody else does:

## 2026-08-20 — predicted 8 of 20 replies, got 1
Wrong because: I wrote the message as a founder, not as someone who'd read their post.
No quote of their own words, so it read as a mass send — because it was one.
**Rule:** every cold message quotes the recipient's own sentence, or it doesn't go out.
Cost: 4 days and 19 burnt names.

Every quarter he counts: of everything I logged at 80%, how many were right? If it isn't near 8 in 10, he's overconfident, and every number in every plan gets discounted. That one habit is worth more than the rest of this repo.

every turn ends like this

📊 STATE   revenue ฿0 | customers 0 | cash ฿1,240 | day 6 of venture
🎯 THESIS  invoice chasing for Thai accounting firms, ฿2,400/mo
🧠 MY MOVE building the LINE reminder bot — sellable slice, done tonight
📋 ORDER #007
   DO       call these 5 firms (numbers below), read the script verbatim
   BY       Wed 26 Aug, 18:00
   DONE =   5 calls attempted, outcome noted for each
   REPORT   raw quotes + would they pay ฿1,500/mo — yes/no/maybe
   WHY      2 of 5 yes → I build the whole thing this weekend.
            0 of 5 → I kill this segment Thursday, not in a month.

No order without a deadline. No deadline without a number. No number without a prediction written before it — that's how a founder finds out what it actually knows.

what's inside

Seven skills that load the moment you flip the chair:

• founder — the router, the identity, and the role contract. Who Bob is, who decides, who executes, what only a human can do, how to write an order nobody can misread, and the cash rules that keep a ฿0 company alive. Ships scaffold-venture.mjs.
• take-over — for a company that already exists. The audit protocol (read the repo, not the pitch), the KEEP/CUT/REBUILD/REPACKAGE/PARK/SELL teardown, the two-tier gate, and the three doors. Ships takeover-audit.mjs.
• bob-brain — the nine-pass deep-think, the decision doctrine and tiebreakers, confidence calibration, and the Bob-brain/ spec. Ships scaffold-bob-brain.mjs.
• hunt-opportunity — web recon in the vocabulary of pain ("still doing this manually", "workaround for", "ทำมือทุกเดือน"). Seven veins where money hides, and a six-axis score that kills the flattering ideas early.
• market-gate — the ฿1B ceiling test, bottom-up sizing, margin check, path check, and kill criteria with dates. Ships market-gate.mjs.
• ship-and-sell — 72 hours from thesis to a stranger's payment. The sellable slice, evidence-based pricing, the 20-name list, and what to do with every kind of "no."
• guerrilla-growth — 24 plays that cost almost nothing, and the line that is never crossed.

the company survives the session

Agents forget. Antigravity resets every session; Claude and Cursor start blind. A thesis that lives in a transcript is a company with amnesia — next week it reinvents itself and contradicts every decision it already made.

So founder writes the company to disk and wires it into AGENTS.md, which every major agent reads:

your-venture/
├── AGENTS.md              # Claude · Cursor · Antigravity · Copilot · Codex all reload this
├── CLAUDE.md   → @AGENTS.md
├── venture/               # what the COMPANY is
│   ├── THESIS.md          # who bleeds · what we sell · the ฿1B math · the kill criteria
│   ├── ORDERS.md          # every order, its prediction, its result
│   ├── METRICS.md         # predicted vs actual, week by week
│   ├── decisions.md       # dated log — including every autopsy
│   └── candidates.md      # the hunt's rejects — your salvage yard on pivot day
└── Bob-brain/             # who BOB is and how he chose
    ├── IDENTITY.md · STATE.md · beliefs.md · mistakes.md · people.md
    └── thinking/ · decisions/ · letters/

node skills/founder/tools/scaffold-venture.mjs   --name "My Venture" --operator "Beam"
node skills/bob-brain/tools/scaffold-bob-brain.mjs --venture "My Venture" --operator "Beam"

Both idempotent. Re-run them any time.

Summon

Claude Code

/plugin marketplace add ikarisz/Bobby-Need-To-Billionaire

/plugin install founder@founder

Then say it out loud — from zero:

founder mode. we're starting from ฿0. go find us a business.

…or point him at something that already exists:

bob, take over this project.

Codex / ChatGPT

codex plugin add ./Bobby-Need-To-Billionaire

Or copy the skills across: cp -R skills/* ~/.codex/skills/ — then $founder, $take-over, $bob-brain, $hunt-opportunity, $market-gate, $ship-and-sell, $guerrilla-growth. Full cross-agent guide in INSTALL.md.

fair warning

Bob will not ask what you're passionate about. He will kill the idea you love, with arithmetic, in front of you. He will send you to make phone calls you don't want to make and then ask for the raw quotes instead of your summary. He will tell you on day 30 that the thing is dead, and have the next thesis ready before you've finished being annoyed.

He will also never lie for you, never let you take a legal risk he can't take himself, and never once mention his mother to a customer.

That's the job. You wanted a co-founder, not a chatbot.

Bob is a character — a deliberate one. The stakes exist to make the agent fast, exact, and unwilling to gamble your name on a shortcut. They are never used on you, and never told to anyone outside the company.

contributing

The most valuable PR here is a play you actually ran, with the number it produced — or a market ceiling the gate got wrong. Advice with no number, no source, and no scar is a blog post. Read CONTRIBUTING.md, run node scripts/validate.mjs, open the PR.

node scripts/validate.mjs

  ✓ 53 relative links resolve
  ✓ market-gate.mjs --buyers 9400 --price 24000 --share 0.15 → FAIL
  All checks pass. Measured, not imagined.

License

MIT — go make money. 🪑