building @vibedraft_app - the Agentic GTM Engine: https://nitter.gq/t.co/T5EMemxynl vibe coding myself to financial freedom. learnings and lessons a long the way.
Boston
Joined December 2023
- Tweets12.4K
- Following3.2K
- Followers2.2K
- Likes12.7K
i feel like github and ci/cd pipelines are becoming a real bottleneck.
if you have many agents making pull requests and waiting on pipelines to run like unit tests and other checks, the time from implementation becomes almost 2-5+ depending on the codebase.
am i just doing it wrong or is there a better way to deploy and merge through version control and automated testing?
context switching is the tax nobody budgets for when they start building after a 9-5.
every time attention moves from a day job meeting to a side project bug, some of it never comes back.
the fix isn't more hours. it's fewer handoffs: batch the work, write the spec once, let an agent hold the context you can't.
build the system so it doesn't need you checking in constantly.
perfect code with nobody watching doesn't convert into a single customer.
rough code with an audience already paying attention does, every time.
founders treat the product as the asset to protect and the audience as an afterthought. it should be the other way around.
polish is decoration. the audience is the distribution.
showing up every day beats waiting for one post to go viral.
impressions jumped 92% in a single week, 473 to 910, sept 15 to sept 21, and nothing in that stretch was remarkable on its own.
the only variable that moved was frequency.
consistency compounds quietly. virality is a lottery ticket you can't control when it pays out.
Ricky retweeted
We'll be spending a lot more time trying to understand the outputs of language models. A few thoughts, tips & tricks:
Writing. Something I've had success with: Ask your LLM to explain something in ASD-STE100, it's a controlled language specification originally developed for aerospace maintenance documentation. LLMs well-versed in this language and it comes with heavy constraints on clean writing style that I often find a lot more readable. Sometimes I've tried to soften it a bit e.g. ask for "80% of the way to ASD-STE100" because the spec is quite stringent. But even better:
Diagrams / images. Instead of writing, ask your LLM to create a diagram. These can be a lot easier to process, parse, and understand. But even better:
Web pages. Ask for output "in HTML" to get a beautiful, interactive webpage. LLMs are getting really good at frontend and can create beautiful experiences, animations, etc. But even better:
Explainer videos. The output format I am most bullish on is fully custom / bespoke explainer videos generated on any arbitrary topic. Experiment with things like "Create a 3b1b style video explainer on X. Use my ElevenLabs API key for audio narration". (you'd need an API key for the latter or you can ask your LLM to find you decent free alternatives that use your local compute). This is actually starting to work!
In summary:
- As LLMs get better, they will do more and more of the legwork autonomously, and a lot more of our work will rise up the abstractions into oversight and understanding.
- Luckily, LLMs can help here too because as intelligence and code are increasingly abundant, you can ask for large, custom, discardable software artifacts (e.g. web apps, video explainers) that would have never made sense to create before. Push the boundaries here and you'll be surprised.
Ricky retweeted
posted 6,000+ tweets since 2020
built an X growth tool and sold it for $10m
so I pulled my own data to see what actually works on X right now
short list 👇
1. replies beat posts when you're small. your post gets 50 views, a good reply on a viral post can get 50k
2. long beats short, my posts over 600 characters crush my one-liners
3. put a real dollar amount in it, those are my best performers
4. one-liners are lottery tickets, a few go mega viral, most die
5. don't ask in the hook. declare. my question hooks underperform
6. if it flops in 30 minutes, tweak the hook and repost. one of mine went from 7 likes to 380
7. people bookmark what they'll use later, my most bookmarked post (19,069) was a guide I shared, with notes
(bookmark this post btw)
8. talk about your life more than your product. posts naming my products: 77 likes median. posts about my kids, the chalet, the plane: 167
on top of that, one core rule: be consistent, never miss a day
this is a hard and long journey, you won't win quick, but it's incredibly rewarding if you put in the effort
treat an agent spec like onboarding a new hire who's never seen your codebase.
- say what success looks like in one sentence, not five
- name exactly which files or functions it's allowed to touch
- hand it the edge case that broke last time, not just the happy path
- tell it how you'll verify the work, not just how you'll know it ran
skip any of these and you'll spend more time untangling the output than you would've spent writing it yourself.
i pulled the waitlist number against the dec 31 target: 2 signups as of nov 4, 50 needed by year end.
a straight line from here puts me behind pace, no hiding from that math.
the real question isn't the product. it's which channel actually closes that gap before the deadline.
cold dms, replies in real threads, or just posting more like this: which one would you bet on first?
profile clicks jumped 68% in a week, 32 vs 19, oct 19 vs oct 12.
impressions don't pay rent. someone clicking through to see who's behind the account is the first real signal they're curious enough to find out more.
chase that number, not the vanity one.
the person who needs what you built doesn't know your name yet.
they're typing their actual problem into a search bar, frustrated, in plain words, no jargon.
find that exact thread and answer it straight, no link attached.
mention the thing you built only after the answer already helped them.
i'm at 2 waitlist signups as of nov 4 against a goal of 50 by year end, and the math doesn't care how good the product is.
talking to 125 founders about their first customer, the fastest growers skipped the landing page pitch and just replied inside real threads where people were already stuck.
if you want your first users, go answer the question before you ask for the click.
impressions went from 473 to 910 in a week, sept 15 to sept 21, up 92%, with no viral post anywhere in there.
the only thing that changed was showing up daily instead of waiting for inspiration to hit.
consistency compounds quieter than any single post ever will.
i watched short link clicks jump 41% week over week, 24 versus 17, oct 12 vs oct 5, and nothing about the product changed that week.
the only variable that moved was how often i showed up and pointed people at it.
distribution is a multiplier on attention you already have, not something you bolt on after launch.
if you're hunting for your first buyers, skip the indie hacker subreddits. that room is builders pitching other builders.
go where people just shipped and are still raw about what's broken:
- product hunt comments on today's launches, makers asking for feedback with zero users yet
- show hn threads, same crowd, more technical, blunter about what isn't working
- category review sites, where the 2 and 3 star reviews are people who still want the category, just not that product
read for the admission, not the pitch. drop the same link in both rooms and only one of them replies.
i interviewed 125 founders about how they landed early customers.
the ones who grew fastest didn't send cold traffic to a landing page, they replied to real threads and let people click through on their own.
a landing page converts strangers once they're already curious. a reply converts someone still deciding.
which one actually gets you a first paying customer: the perfect landing page, or the reply that meets someone mid problem?
code can be rewritten in a few months. trust takes years and can't be cloned.
most founders protect their codebase like it's the asset and give away their audience's trust for free.
that's backwards.
profile clicks move before followers move, and followers move before signups move.
mine jumped 68% week over week, 32 versus 19, as of 2026-10-19, with no viral post behind it.
quiet metric, real signal.
launched with zero customers, zero audience, zero budget.
what's the one channel and approach you'd bet on first to find real buyers, not just eyeballs?