The Boring Developer

2 to 1,000 followers on X in 81 days, mostly by replying. july: 5,115 impressions. august: 156,854. so i built humanish. it finds posts worth replying to and drafts in your voice. you press send. $29/mo: tryhumanish.com/?utm_source=…

i think a model refusal should never look like a short result. in novakit it used to pass as a short listing. now i handle refusals on purpose and tell the customer what happened and what to add

google approved reviewsaathi for the business profile api after i applied twice. it reads reviews and posts approved replies on google itself now. still no idea how to use the waiting in between

Looking to #connect with n8n workflows folks: 🧩 sub-workflows (spaghetti, tidy) 🔐 credentials vault (shhh) 📬 gmail triggers (inbox chaos) 🧾 invoice bots (boring, good) 🪵 execution logs (read at 2am) 🧱 http request nodes (headers, ugh) if you build these, let's connect 🤝

before i trust a drafted reply i ask who reads it first. in ReviewSaathi it's the owner: every draft waits for approval, and 1 to 3 star reviews always go to them first. it stuck because the owner knows things about the customer that i don't

my guess: a review will be the first place someone tries to boss a business's AI around. instructions hidden in a Google review are just data to ReviewSaathi, it ignores them. no proof anyone is trying yet, so i could be early

i think patient feedback gets given where the person already has the app open. CareLoop collects it on WhatsApp and Google reviews, handles complaints and sends the owner one report. no data on the first part yet, just a hunch

a beginner building tools for small businesses would sell the tool. a veteran would sell the boring job it takes over. ReviewSaathi replies to every review and sends the owner a short report. i lean to the job, though i'm no veteran

how to find X posts worth replying to (the 3 filters i use) 2 to 1,000 followers in 81 days, mostly by replying. these are the filters that decided which posts got a reply. the play: right accounts > fresh posts > one specific point > first steps - go to: x.com/search-advanced - put your topic in "these words". pick 2 or 3 topics you can actually say something about - under engagement, set minimum likes to 10, so you land on posts people are already reading - set "from" to yesterday's date. search, then open the Latest tab > the 3 filters - the account has 5,000+ followers, unless you two already talk. hover the name to check - the post is under a day old. after that, everyone has scrolled past - you have one specific thing to add: a number, a step, a mistake you made. if all you have is "great point", skip it > the catch - most replies stay quiet. 15 to 27 oct, my middle reply got 17 views. my best got 1,349 - per post, my own posts did better: 168 views vs 49 for a reply. replies are how new people find you, not a shortcut - one account's numbers, not a rule from X > what not to do - don't reply to the same person twice in a day - don't paste an AI draft as it is. people spot it (someone once answered mine with "AI reply????") - don't put your link in a reply i built humanish to run these filters and draft the reply in my voice. i still read and send every one: tryhumanish.com/?utm_source=… reply with your topic and i'll tell you which filter i'd tighten first. bookmark this for your next reply session

a duplicated line broke the CareLoop admin login. the syntax check passed, the page didn't load. now every page gets opened in a headless browser before i say it's fixed

how do you thank a clinic's reviewer without hinting they were a patient? ReviewSaathi's clinic replies never mention a treatment, a condition or the patient part. a draft that does gets rewritten, then falls back to a safe short line. i can't tell if the safe line reads too cold

for what a small tool should do first, i pick one exact job. whatismyage.me works out your exact age and nothing else. my guess is the single job is why anyone would remember it

a one-click signed fake whatsapp webhook should exist. for every CareLoop test i post one from the server, with fake patient numbers and rolled-back SQL, then delete every test row after

Looking to #connect with AI agents and automation folks: 🤖 AI agents (hot takes only) ⚙️ Automation (nerding out) 🧠 Solo founders (low key) 🚀 Build in public (come argue) 📈 Bootstrapped SaaS (lurking less) 🔍 Context engineering (ask me) if that's you, let's connect 🤝

changed one thing in my review replies: they thank the person by first name and mention one detail from the review, in two or three sentences. CareLoop gets the opposite rule, no doctor names, diagnoses or departments in a patient message. same job, different line to hold

my guess: three reminders is about where chasing stops helping. CareLoop chases an open complaint every 6 to 24 hours depending on priority, at most three times, then the owner gets a backstop alert. i could be wrong about three

imagine a glowing review that also mentions a refund. ReviewSaathi sends that one to the owner even with high stars, along with legal threats, safety and accusations against staff. i think i'd want to see it before any reply goes up

i'd guess more video gets built like web pages than like timelines. every episode of my youtube show is a set of pages rendered to video with HyperFrames, and i haven't opened an editing timeline for it

a beginner in salon automation would build an app for the owner to learn. a veteran would use the chat she already has open. Rebound runs a salon owner's day from one WhatsApp chat, bookings to reminders to attendance. i lean chat

before i automate anything i ask if it's rare, under 5 minutes and harmless when it breaks. if all three, it stays by hand. ReviewSaathi runs its own small gate on drafted replies, right name and no duplicates, and retries one that fails

the scary button is the one you can't undo. ReviewSaathi's Google posting is a dry run by default. it shows what would go up, and only a second confirm actually posts