@Merisdabhi: signal 55/100
Is this your account? Share your reading:
#30 of 190 accounts read, fewer needle hits than 84% of them. The leaderboard
Most of these 18 posts are about other people's work. With most of them about what @Merisdabhi builds, measures or lives, this account joins Find, where people look for builders worth following. The next reading counts the newest posts.
Meris Dabhi, the last 18 posts, read by six needles; the score counts the 7 that are words only. A reading of the words, not a verdict on the person, and not a prediction of reach.
Every post, ranked
Best to weakest by overall signal. Pick a needle to rank the posts by that needle alone.
-
1 of 18 Strongest post
54views2likes0reposts1replies
Jev gets just 2 choices in Flappy Bird. What should an AI know before pressing either one? 🐤 FLAP → go up WAIT → keep falling I’m working on making Jev control the bird through real API decisions. Here’s the setup for developers building real-time AI agents: → 8 inputs for every decision Bird Y, vertical velocity, next pipe X, pipe distance, gap center, gap size, vertical distance to the center, and estimated arrival time. → Predict the next moment The bird can be aligned with the gap and still be falling too fast. Jev needs to answer: “Where will I be when I reach that pipe?” → Target decisions every 50–150 ms Roughly 7–20 decisions per second, when latency allows. Only 1 request in flight at a time. 60 FPS rendering shouldn’t mean 60 API calls per second. → Execute Jev’s answer directly Jev returns FLAP → the bird flaps. Jev returns WAIT → gravity continues. → Make every decision visible 6 logged fields: action, bird Y, velocity, pipe distance, gap center, and latency. A debug overlay draws the predicted trajectory and target point. The experiment: can better state information and decision timing turn Jev into a reliable player? Want the full build prompt? Comment “JEV” → you must be following + repost to receive it free.
This post carries a picture or video. The needles only read the words, so it is left out of the account's score.
signalVague1%Nothing new35%Secondhand8%Nothing to use8%Bait or brag97%Hype13% -
2 of 18 1.8Kviews31likes2reposts30replies
I got 5 paying customers in 3 days with basically ZERO audience. No big ad budget. No YouTube. No Instagram. No existing audience. Just → X + Reddit. Here’s exactly what happened: → I launched my app → I had ~370 Reddit followers + 26K+ karma → I learned I could about marketing on Reddit → I posted my app in a few relevant communities → 1,200+ people visited my app → 5 actually became paying customers And this point I felt it's very easy to grow app😅 I knew Reddit was strict about self-promotion. But there’s one thing I love about Reddit: → If you get the right post in front of the right community, you can get thousands of visitors without spending money. Then... I got banned. 💀 And now my MRR is basically stuck. That experience taught me something: Building the app wasn't the hardest part. Distribution was. So I'm curious: → How are you marketing your app? → What distribution channel is actually working for you right now?
This post carries a picture or video. The needles only read the words, so it is left out of the account's score.
signalVague1%Nothing new67%Secondhand3%Nothing to use14%Bait or brag74%Hype31% -
3 of 18 155views5likes0reposts2replies
I got 5 paying customers in 3 days with basically ZERO audience. No big ad budget. No YouTube. No Instagram. No existing audience. Just → X + Reddit. Here’s exactly what happened: → I launched my app → I had ~370 Reddit followers + 26K+ karma → I learned I could about marketing on Reddit → I posted my app in a few relevant communities → 1,200+ people visited my app → 5 actually became paying customers And this point I felt it's very easy to grow app😅 I knew Reddit was strict about self-promotion. But there’s one thing I love about Reddit: → If you get the right post in front of the right community, you can get thousands of visitors without spending money. Then... I got banned. 💀 And now my MRR is basically stuck. That experience taught me something: Building the app wasn't the hardest part. Distribution was. So I'm curious: → How are you marketing your app? → What distribution channel is actually working for you right now?
This post carries a picture or video. The needles only read the words, so it is left out of the account's score.
signalVague1%Nothing new68%Secondhand3%Nothing to use13%Bait or brag74%Hype31% -
4 of 18 67views2likes0reposts0replies
I’m fine spending Astra or Opus 5.5 usage on a difficult bug. Spending it on “find the file” is harder to justify. That’s what stuck with me from a breakdown of Astra usage tips. You ask for one change. The agent searches the workspace → opens a pile of files → makes the edit → writes an essay explaining it. Some exploration is necessary. But if you already know the file, why make it hunt? Before buying more usage, I’d start here: → Know the file? Paste the exact path. → Know the scope? Say what needs changing. → Working in a big repo? Keep a short workspace map. → Need a quick answer? Ask for one. → Have a routine task? Try a cheaper model first. For bigger jobs, I’d test the video’s “Mind and Hand” approach: Astra or Opus → diagnosis and plan Cheaper model → clearly defined edits Tests → check the result Astra or Opus → review the difficult parts The catch is the handoff. “Fix this” leaves the next model guessing. “Edit these two files, preserve this behavior, and pass these checks” gives it something concrete to finish. If the stronger model has to redo everything, the savings disappear. I want to spend my quota on the bug I couldn’t solve. I can paste the file path myself. What’s one task you’ve stopped giving your strongest model?
signalVague4%Nothing new49%Secondhand31%Nothing to use3%Bait or brag48%Hype5% -
5 of 18 148views3likes0reposts1replies
I don’t want another $29/month AI subscription. But I’m also tired of becoming tech support for my own “free” tools. I tried setting up open-source AI to save money. Between the errors, crashes, and updates, I kept losing time I wanted to spend on my ideas. With releases like Qwen-Image-2.1, alongside downloadable models from DeepSeek and Z ai’s GLM family, I understand the excitement around running AI yourself. I’ve asked the same question everyone else asks: “Why pay for a subscription when there are free alternatives?” So I tried setting up open-source tools, local models, and platforms myself. Getting them running was only the beginning. After that came managing them, troubleshooting errors, and keeping everything updated. The frustrating part is having an idea you’re excited to try, then spending that time fixing the tool you wanted to use. And even after fixing it, you have to get your attention back to the original idea. That’s what changed how I think about subscriptions. The monthly price is easy to see. The time spent maintaining your own setup is easier to overlook. Open source can be worth the effort. Control, customization, and learning how things work all have value. But right now, I want to spend more time testing ideas, improving my process, and creating things. If a paid service helps me do that with fewer interruptions, paying starts to make sense. It could even be a service built on an open model. I still care about the subscription price. I’m just starting to count my time, too.
signalVague2%Nothing new69%Secondhand6%Nothing to use11%Bait or brag9%Hype6% -
6 of 18 188views6likes0reposts6replies
wait a minute... i just re-subscribed to claude to use opus 5.5 and they gave me $100 in free credits 💀 what is going on 😭
This post carries a picture or video. The needles only read the words, so it is left out of the account's score.
signalVague3%Nothing new50%Secondhand7%Nothing to use46%Bait or brag36%Hype10% -
7 of 18 80views0likes0reposts1replies
Apollo → ~$0.01/lead Hunter .io → ~$0.02/lead Clay → ~$0.01–$0.10+/lead Now imagine a fourth option: → Email finding from $0.0089 per result → ZERO monthly subscription That’s Treg. Connect it to Claude Code or Codex and describe the contacts you’re looking for in plain English: “Find marketing leaders at Series A SaaS companies and look up their work emails.” Your agent searches across providers to find the people and contact details you need. Pricing varies by provider and task. → Add funds → Describe your ideal contacts → Run your search → Pay as you go Need leads for a one-off campaign? Run it. Need more next month? Come back then. → No monthly plan to keep paying for between campaigns → No yearly commitment to get started The price gets your attention. The freedom from another subscription is why you take a closer look.
This post carries a picture or video. The needles only read the words, so it is left out of the account's score.
signalVague1%Nothing new43%Secondhand69%Nothing to use13%Bait or brag24%Hype32% -
8 of 18 435views2likes0reposts1replies
13 AI releases, previews and specialized updates in 9 days. Even if you test 2 a day, you need almost a week just to give each one a first look. Before you’ve picked a favourite, there’s another launch. Here’s the list for 15–23 September ↓ 1 → Claude Opus 5.5 Anthropic · 22 Sep 2 → GPT-6 Sol OpenAI · 22 Sep 3 → GPT-6 Luna OpenAI · 22 Sep 4 → Grok 4.7 xAI · 21 Sep 5 → MiMo-V2.6-Pro Xiaomi · 21–22 Sep 6 → MiMo-V2.6-Flash Xiaomi · 21–22 Sep 7 → Qwen-Image-2.1 Alibaba · 20 Sep 8 → Step 5 Preview StepFun · 20 Sep 9 → Qwen3.8-Omni-Flash Alibaba · 18 Sep 10 → Grok Voice Transcribe 2.0 xAI · 18 Sep 11 → Astra for Law OpenAI · 17 Sep GPT-6 Astra configured for legal work. 12 → Gemini 3.8 Live Google · 15 Sep 13 → Gemini 3.8 Live Extended Thinking Google · 15 Sep And the product updates keep coming: → Google Search Live powered by Gemini 3.8 Live. → Intrinsic Core, an open-source robotics platform. → Meta One, with expanded AI and business tools. → Reported Muse expansion to Messenger / Telegram. → WhatsApp Business AI agent tools to track; release timing unconfirmed. This is where I think of the woman in the red dress from The Matrix. Another launch catches your eye. You stop what you’re doing. Suddenly, you’re testing another tool while your actual project sits untouched. Stay focused. → Pick a problem. → Choose a model that handles it. → Finish something useful. → Test a replacement when it could solve a specific limitation. You don’t need to try every model to build something valuable. Give each new release a reason to interrupt your work.
This post carries a picture or video. The needles only read the words, so it is left out of the account's score.
signalVague1%Nothing new40%Secondhand90%Nothing to use4%Bait or brag15%Hype16% -
9 of 18 117views1likes0reposts0replies
Your next AI business idea might be hiding in someone’s revision requests. Google just shared a useful example from New York Fashion Week. Its engineers worked with designers Jane Wade and Sergio Hudson to build specialized tools in Google Flow. One helped assemble runway looks before producing physical samples. Another let a designer explore lighting, props and runway layouts within his budget. That second example is worth paying attention to. Changing a creative decision can trigger another rendering, another handoff, another round of work. A tool that makes those revisions easier has a specific job someone already needs done. If you’re looking for something to build, ask a working professional: “What change does your client keep requesting that takes far too long to make?” Then find out: • What do they have to redo? • Who else gets pulled into the revision? • How do they judge whether the result is usable? Build around one repeated frustration. Test it on a real revision. “AI for creators” is a huge category. “Preview the client’s lighting changes before commissioning another render” gives you something concrete to test.
signalVague2%Nothing new43%Secondhand96%Nothing to use4%Bait or brag10%Hype18% -
10 of 18 88views2likes0reposts0replies
That “me at 3am” video? You can make one without being awake at 3am. Or filming yourself at all. The videos you shared have a simple structure: → An everyday scene → A small reaction → A caption that feels a little too personal “Building another feature when nobody understands the first one.” Every developer knows that feeling. You can create videos in this style with AI: → GPT-6 Astra writes the situations and captions → Higgsfield generates the character and scene → You pick the take where the reaction fits the joke No camera setup. No recording your face or voice. Start with something your audience has actually experienced. What’s the thought they’ve had a hundred times but never posted?
This post carries a picture or video. The needles only read the words, so it is left out of the account's score.
signalVague3%Nothing new53%Secondhand89%Nothing to use5%Bait or brag39%Hype10% -
11 of 18 32views1likes0reposts0replies
A $500/month AI service needs 10 customers to reach $5,000/month in revenue. The hard part is defining a job worth paying $500 for every month. I’d start with a boring task a business already pays someone to finish. → One customer type. → One repeatable job. → One clear definition of “done.” For example: turning a supplier’s messy product spreadsheet into a checked catalog, ready to upload. An illustrative offer: → $500/month. → Up to 500 product records. → One catalog update per month. → Human review before delivery. The customer buys a finished catalog. You own the process that produces it. Here’s how I’d test it: 1 → Find 3 paying customers with the same recurring problem. 2 → Use AI to do the work and check every output. 3 → Record every error, missing input and revision request. 4 → Turn those lessons into instructions and quality checks. 5 → Standardize the process before adding more customers. The error log deserves more attention. “Clean this spreadsheet” leaves a lot undefined. A useful rulebook says: → Preserve every SKU. → Flag missing dimensions. → Reject duplicate entries. → Never invent a specification. Then measure the entire job: Inputs → AI processing → Review → Corrections → Delivery. At the example price: → 3 customers = $1,500/month. → 10 customers = $5,000/month. → 20 customers = $10,000/month. That’s revenue. Your costs include tools, review, revisions and customer acquisition. Start by proving you can deliver one clearly defined outcome reliably and profitably. Then earn the next customer.
signalVague1%Nothing new64%Secondhand87%Nothing to use3%Bait or brag7%Hype4% -
12 of 18 131views2likes0reposts1replies
1 cold email → $250,000 deal. Sounds fake. Until you see how the guy behind it operates. He sends 1M+ cold emails every month. He’s booked 3,000+ sales calls through outbound. And he recently closed a $250K deal from a single cold email. The crazy part? He didn't use a “free audit.” He says that offer is so overused that prospects immediately put their guard up. Instead, his email had ONE job: sell the next step. He broke down the entire system: → The exact $250K email, line by line → 3 alternatives to the “free audit” → His guarantee/offer formula → 5 Claude-generated ad variations → His follow-up framework And he's giving the whole thing away for free. Comment “DEAL” and connect with him for the link. Save this. You’ll want it when you write your next outbound campaign.
This post carries a quoted post. The needles only read the words, so it is left out of the account's score.
signalVague1%Nothing new52%Secondhand93%Nothing to use28%Bait or brag98%Hype82% -
13 of 18 280views1likes0reposts0replies
this week felt boring. now there are 7 models from 5 AI labs on the watchlist. here’s what’s actually behind the hype: → Anthropic: Sonnet 5.2, Opus 5.2, Fable 5.2 testers are reporting possible new versions running behind existing Claude options. interesting signals, but the names and release dates remain unconfirmed. → Google: Gemini 4 Pro suspected Arena tests are circulating. the demos have people talking, but identifying an unreleased model from its output is still a guess. → xAI: Grok 4.7 Elon has publicly discussed it and delayed it for more work. this one has a direct signal from the company’s founder, though the timing keeps moving. → OpenAI: GPT-6 Sol? Sam pushed an unnamed launch to next week. Tibo has hinted at Tuesday. the missing detail: neither teaser identifies it as GPT-6 Sol. → Moonshot: Kimi K3.1? a reported cryptic Kimi post has people decoding a possible 3.1 hint. intriguing, still speculative. that’s the watchlist heading into next week. even two or three of these landing would give us plenty to test. which one has the best chance of replacing your daily model?
signalVague2%Nothing new47%Secondhand89%Nothing to use39%Bait or brag47%Hype16% -
14 of 18 138views0likes0reposts0replies
Perplexity subscribers, what’s stopping you from cancelling? Look at what’s already available: → ChatGPT can search and do deep research → Claude can search and research → Gemini has Deep Research → Grok has DeepSearch Now Perplexity is adding video generation. → MiniMax’s models → ByteDance’s models → Another feature powered by someone else’s tech Having everything in one place has value. But if you already pay for one of these AI tools, what makes Perplexity worth paying for too? Better research? A better workflow? What’s the one thing keeping you subscribed?🤔
This post carries a quoted post. The needles only read the words, so it is left out of the account's score.
signalVague3%Nothing new79%Secondhand97%Nothing to use42%Bait or brag62%Hype9% -
15 of 18 33views2likes0reposts1replies
I have a free reset, too.
This post carries a picture or video. The needles only read the words, so it is left out of the account's score.
signalVague41%Nothing new92%Secondhand35%Nothing to use86%Bait or brag28%Hype3% -
16 of 18 797views27likes0reposts31replies
hey everyone 👋 i’m a solo founder looking to connect with more builders working on: → saas → ai → dev → automation → indie hacking → marketing building in public is better with the right people say hi and let’s connect.
signalVague37%Nothing new91%Secondhand72%Nothing to use64%Bait or brag61%Hype12% -
17 of 18 60views1likes0reposts1replies
✅ AI writes text ✅ AI generates & edits images ✅ AI generates music ✅ AI writes code 🎯 Next → AI edits entire videos
This post carries a picture or video. The needles only read the words, so it is left out of the account's score.
signalVague32%Nothing new86%Secondhand93%Nothing to use76%Bait or brag15%Hype46% -
18 of 18 Weakest post
1.1Kviews32likes0reposts61replies
hey 👋 @X looking to connect with people building things. would love to meet: → indie hackers → founders → SaaS builders → developers → indie makers → AI builders if you’re building something too, Drop a “hi” 👋 and let’s connect.
signalVague87%Nothing new94%Secondhand97%Nothing to use74%Bait or brag77%Hype2%
Read on 2026-09-25. This is your account and you want it gone? Write to hello@postneedle.com. The leaderboard