claude code is vibe coding. bevel is vibe producting.

a chat that watches your product 24/7, decides what to ship next, and hands you the claude code prompt to do it.

55 visitors, 1 signup. the auth flow eats every distribution push.
two founders shipped adjacent tools on hn. dm them while it's warm.
your icp is on indie hackers saying distribution is their #1 problem.
the new stack

claude code is vibe coding. bevel is vibe producting.

vibe coding

you say "build me X". claude code writes the code.

claude code, cursor, codex. you type the change, the agent runs the keystrokes. solves how. assumes you already know what to ship.

vibe producting

bevel watches your product. tells you what to ship. hands you the prompt.

one chat watches posthog, github, vercel, and the open web around the clock. writes the next ticket. drafts the claude code prompt. you decide what's worth shipping; bevel handles the rest.

write. ship. the path.

writes the next ticket. 3 sharp ones, every scan.

every 12h, bevel writes 3 things worth shipping: a launch tweet, a cold dm to the 8 users who hit cmd+k, the code change. stage-aware, so a pre-pmf founder gets distribution, not bug fixes nobody hits yet.

autopilot scanning
tickets
raise free cap 250 → 1k issues teams bounce after 7d at cap. lever.
START DISMISS
dm the 8 power users who hit cmd+k they tried to summon nothing. ask why.
START DISMISS
audit linear pricing drop $29 → $9. you need a story.
START DISMISS

bevel writes new ones as it scans your product + reads your chats.

quick actions
  • interview a recent signup
  • draft this week's update

hands you the prompt. paste into claude code. ship.

bevel writes claude code prompts, not code. each names the file and the exact change. paste, run, ship. mark it shipped and bevel checks back the next day to see if your numbers moved.

bevel

raising the free-tier cap from 250 → 1,000 issues. evidence is consistent and the prior raise in wk 9 paid off; this one should too. paste the prompt into claude code.

claude code prompt · 4 lines
# bevel · ticket · paste into claude code
raise free-tier issue cap 250 → 1,000
update src/app/pricing/page.tsx to match
fire the in-app warning at 80% of new cap
add a regression test: cap_hit fires at 1,000

pick a direction. bevel re-tunes around it.

one knob that moves everything else. commit the week to distribution, launch, or polish. every ticket and chat answer re-tunes to advance it. one bet at a time, not a backlog of 40.

the path day 3 of 10
distribution-first active

land your first 50 founders this week. stop everything else.

skips: shipping, polish. founder-driven sales is the whole job.

expansion bevel's lean

go public. show hn, product hunt, twitter. spike awareness.

skips: conversion. you'll get visitors who bounce.

product perfection

no users yet = no pressure. polish until the next visitor becomes a fan.

skips: distribution. nobody sees this work until you switch back.

switch any time. tickets re-tune within the hour.

what makes the loop work

the loop is the moat. three reasons.

ai recs are easy. an ai that watches, decides, ships, then comes back to ask if it worked, that's the loop. three things make it hard to copy.

01 real names

20 prospects. not vibes.

ask "find me 20 founders to interview" and bevel pulls real named people from hn, reddit, and linkedin, each with the post you can open. nothing fabricated.

  • jane · "i'd pay for this"hn
  • marcus · 87 upvotesr/saas
  • priya · launch · 12dhn
02 the loop

comes back to ask if it worked.

click ship and bevel remembers your numbers. queues a check-in: 3 days for outreach, 14 for pricing copy, 30 for retention. the next ticket asks whether it moved.

check-in · 14 days

"did the cap raise move activation?"

03 every move

not just code shipped.

track impact fires on outreach, landing copy, pricing experiments. anything you actually go do counts. five verbs in, five verbs tracked.

code outreach post research watch

you slept. bevel scanned. here's what it found.

fix the auth flow eating my signups
bevel
pulling posthog · last 30 days
reading the signup_complete + /login funnel
checking src/app/login

found a signup leak in your sleep.

  • 659 pageviews, 55 unique visitors, 128 sessions in 30 days
  • the may 19-20 dau spike (show hn) sent real traffic to the landing
  • but signup_complete fired once. /login: 49 views, 6 uniques, 1 completed
  • the cognito auth flow is where every distribution push dies

here's the prompt to ship the fix:

claude code prompt 16 lines
# bevel · ticket · paste into claude code

look at src/app/login (the cognito signup flow).
find the highest-effort step between the landing CTA
click and the authenticated /app redirect: email
confirm, captcha, extra fields. remove or defer it.

goal: click "start the autopilot" on the landing,
enter email + password, land in /app. zero interruption.

if cognito requires email confirmation, add a
resend-confirm banner to /app so anyone who bounced
mid-flow can recover.

don't refactor anything outside the auth path.
run pnpm typecheck && pnpm lint after changes.
usage

cursor is the HOW. bevel is the WHAT.

 
bevel the WHAT
claude code the HOW
posthog ai dashboard ai
linear / asana backlog
writes tickets across code, posts, outreach, research
·
·
·
watches posthog + github + vercel + the open web 24/7
·
partial
·
hands you a paste-ready claude code prompt
·
·
·
checks back the next day to see if your numbers moved
·
·
·
memory across decisions; learns your taste
·
·
·

swipe to compare →

pricing

solo beta · free
$0 while in beta

the cockpit. one source, soft caps, full autopilot.

  • autopilot · always on
  • connect 1 source (posthog, github, or vercel)
  • 30 manual scans / month
  • memory across every chat
start the autopilot
team coming soon
$99 /mo

2-5 founders sharing one product memory.

  • shared cockpit · shared memory
  • custom sources · MCP
  • direct line to the maker
available after beta