Sabitlenmiş Tweet
English
Ramanjit Singh
1.4K posts
Ramanjit Singh
@ItzRamanCoder
Building Luna - the AI Chief of Staff that remembers your life and tells you what you're about to drop. DM for Beta.
Remote Katılım Kasım 2018
94 Takip Edilen48 Takipçiler
i can design a backend that deletes three services and runs the whole thing on one loop.
i cannot yet get a single stranger to care that it exists.
turns out those are completely unrelated skills, and only one of them is on my resume. this month is just me learning the second one in public, badly, out loud.
English
solo founders. drop your real follower count. no context, no shame.
i'll start: 20.
curious how many of us are actually building from zero right now, vs how it looks on the timeline.
English
"agents need to be personalized. they need to learn and grow with the user. they need memory."
this is the whole thesis and it's finally uncontroversial.
the interesting fight now isn't whether agents need memory. it's who gets trusted enough to hold the memory of your actual life. that's a trust problem long before it's a tech one.
x.com/supermemory/st…
supermemory@supermemory
Supermemory has raised $3M in pre-seed funding to build the best memory engine for agents. Agents needs to be personalized. They need to learn and grow with the user. They need memory. And that's the future we're building today. x.com/DhravyaShah/st…
English
the quiet shift in AI this year that won't make a single slide.
the best setups aren't one giant model doing everything. it's a cheap model doing the actual work and an expensive one just checking it now and then.
intelligence is getting unbundled into tiers. the teams who figure out the tiering will quietly beat the teams still just buying the biggest model.
English
killed three services this month. one collected data, one ran the logic, one aggregated it all before anything else could touch it.
replaced the whole chain with one loop that just queries the database directly when it needs something.
felt like a demotion for about a day. less code, less that looked like real architecture. then it started working better than the three-service version ever did, because the thing reading the data is the same thing deciding what to do with it.
English
"added a shared whiteboard, haptics, 5 new mini games, and like 25 other things you'll just have to open the app to see."
this is the cadence i keep trying to learn from.
ship a pile of small real things, tell people plainly, repeat. no grand relaunch, no manifesto. just visible momentum, every single week.
x.com/jackfriks/stat…
jack friks@jackfriks
new @lovelee_app update is live i added a shared whiteboard, sound effects, haptic feedback when playing with your pet piggy, 5 new multiplayer mini games to play with your partner and like 25 other things you’ll just have to open your app to see :)
English
English
the month after launch is scarier than launch day and nobody warns you.
the adrenaline's gone, the number isn't moving, and you still have to show up and get one more stranger to care.
that month is the actual test. launch day is just the starting gun.
English
Me and My cofounder are delusional enough that we will beat you.
English
English
the strangest thing i built for Luna is a cron that turns Claude into its own worst customer.
twice a day it wakes up, uses Luna like a busy founder would, reads the brief, plans a day, then grades the whole thing like a chief of staff who expected better.
it's caught more real problems than any single piece of human feedback so far. the machine reviewing the machine, quietly, every morning.
English
"the most interesting insight from the AI avatar experiment wasn't the growth, it was the near-zero backlash."
this tracks with something i keep noticing.
people don't reject AI doing the work. they reject AI pretending it isn't AI. say plainly that it's a bot, make it genuinely useful, and the backlash mostly evaporates.
x.com/rowancheung/st…
Rowan Cheung@rowancheung
An update on my AI avatar that's taken over my social media video content: The account has grown to 48.4k followers and racked up 7 million+ views. But the most interesting insight from this experiment hasn't been the growth, it's been the *unexpected* near-zero backlash. When I started this experiment, I was preparing for the criticism *I thought* I would get for delegating my face + voice to an AI avatar. I disclosed that it was an AI avatar (of myself) everywhere: Tags, descriptions, comments, and even in the videos themselves. But people just don't care, as long as the content delivers value. The truth is, my team and I still spend hours doing research, writing/scripting, and editing these videos. In a nutshell, the workflow is: Newsletter/tweets → Script → AI avatar generation → Edit 1 → tweak avatar → Final edit Ironically, recording myself might be faster (we often regenerate the avatar multiple times to nail certain words). But here's the thing: my avatar is way better than me on camera. And that is my point. The end result is all that matters. As a content creator, you need to do everything in your power to serve your audience in the most impactful way possible. By delegating my face/voice, I can focus on what I'm actually good at – finding interesting stories and writing about them. If you're against using AI, it's like trying to beat someone with a calculator at a math competition.
English
Luna, the honest scoreboard tonight:
followers: 20
waitlist: still 0, the page isn't even live yet
that's on me, not the idea. fifteen days past my own launch date and i keep polishing the product instead of shipping an ugly signup form.
exactly the mistake i'd call out in anyone else in a heartbeat. it goes up this week, ugly or not.
English
"why founders fail: build something nobody wants, don't know how to get customers, give up too early."
i'd add a fourth that quietly eats solo builders alive.
falling in love with the build and calling it progress while zero strangers have seen it. shipping to your own laptop isn't shipping.
x.com/marc_louvion/s…
Marc Lou@marclou
Why startup founders fail: 1. Build something nobody wants 2. Don’t know how to get customers 3. Give up too early And how to fix it: • Build one feature • Ship fast • Kill faster
English
everyone benchmarks how smart the model is.
nobody benchmarks whether it remembers you.
a genius that forgets your whole life every morning is a party trick, not an assistant. the entire game in consumer AI is memory, and almost nobody is actually playing it. that's the bet behind Luna.
English
ran the morning brief today and Luna brought up something i mentioned once, in passing, almost three weeks ago. i never said it again. it did.
i built the thing and it still made me stop and go wait, how did you hold onto that.
every tool, every database call, every line of the loop exists so that one moment lands reliably, twice a day, forever.
what's the last thing an app remembered about you that you'd genuinely forgotten?
English
"claude still feels like a general purpose coworker."
this is the distinction that matters and almost no benchmark captures it.
a coworker isn't the one with the highest score. it's the one who remembers what you were doing last week and picks it back up without a briefing. reliability over raw smarts, every time.
x.com/mckaywrigley/s…
Mckay Wrigley@mckaywrigley
some random ai thoughts: - for code, i went from 80/20 claude/gpt to 80/20 gpt/claude in <3 months. surprised by this tbh, and interested to see where the split is at in another 3mo. - claude still mogs gpt for non-coding agent stuff. codex feels like an engineer (which is great for coding!), whereas claude still feels like a general purpose coworker. gpt still lacks that coworker magic - i’m pretty meh on opus 4.7. my experience hasn’t been *bad*, but it certainly hasn’t been good. sideways if anything. - anthropic has got to figure out the compute thing. you can feel it as a user. vibes are all out of whack bc of it. my opinions above are all likely downstream of this. it’s an issue. - anthropic labs continues to be the goat of ai product. claude design is another hit. it’s fantastic. idk why it’s not talked about more? a+ - updated claude code app is great. i finally switched out of the terminal for it. very well done. - how are people STILL sleeping on the claude agent sdk? i feel like i’m going insane. - gpt 5.5 is incredible. the level to which i trust it for engineering is amazing. if i could only have one model rn, it would be this one just bc of strong need for the coding use case. - codex team is killing it. app has been the gold standard since 5.3 release (buuut i credit conductor team for the ui innovation that everyone is using now). though i could do with a little less passive aggressive shots at ant from the codex team. TARS, dial up class by 30%. it’s a long race guys haha - i uninstalled cursor this month and am now back to vs code for my ide. composer just can’t hang with claude/gpt, and the product feels a bit all over the place. pretty stoked about the xai thing though, because their team is absolutely stacked and i’m excited to see what they might be able to do with that compute. codex and claude code are t1, cursor is t2. i would love if this deal got xai/cursor to t1 for a real trio there. - gemini…? seems like this is 2-3 models now where the model seems like a great release and then nobody ever uses it? i’m bullish google/deepmind but weird it hasn’t translated to product use in any form. kinda disappointed still - no open source models have hit the opus 4.5 level. was hopeful the new deepseek would get there, but nope. good oss agents will have to wait a few more months it would seem…
English
most AI is optimized to talk to you more. more check-ins, more 'just following up,' more notifications dressed up as helpfulness.
i built the opposite into Luna. on a quiet day it should cost almost nothing and say almost nothing.
an assistant that talks constantly isn't attentive. it's just loud.
English
building an AI that reads your calendar and inbox, the scariest line of code is the one that could pull up the wrong person's.
so in Luna every query gets your id stapled on last, after whatever the model asked for. it structurally can't reach into someone else's life even if the model tries.
trust isn't a feature you bolt on once the thing already touches your inbox. it's the first thing, or it's nothing.
English