Sabitlenmiş Tweet
michael s galpert
40K posts
michael s galpert
@msg
See you at the next @clawcon run a product studio that enables people with ai. previously worked on Fortnite and a bunch of startups.
Katılım Mart 2007
5.4K Takip Edilen17.1K Takipçiler
just got off a call with a company building for autonomous agents and my head is spinning
English
when you are building a hammer everything looks like a nail.
me rn
GIF
English
@its_MayYang @clawcon @lionelsimai Thanks so much for all your help making last night possible. You were a fantastic emcee!!
English
michael s galpert retweetledi
Wrapped up @clawcon Singapore yesterday. What an event, with more than 500 participants.
My favorite session was the demo on how to make your OpenClaw feel more human with real personalities. Flashbacks to Westworld.
Thank you @lionelsimai and @msg for trusting me to emcee the event. Looking forward to growing this phenomenon across SEA.
English
michael s galpert retweetledi
English
@lawheroezV2 @tszzl It’s for the same reasons outlined in this sermon
Zohar Atkins@ZoharAtkins
English
michael s galpert retweetledi
The team just released an old idea reimagined for a new reality. A floating dock that shows what your team is shipping and their live coding sessions.
Slashtalk lives at the edge of your screen and quietly aggregates Claude Code session data, GitHub activity, and live work signals — organized around the repos.
Just something fun we thought to put out in case others find it useful.
github.com/HereNotThere/s…
Fei@feiii_____
the hardest part of running a small team now isn't shipping fast - it's keeping eng and product in sync without pinging each other constantly. so to help our team coordinate better, we created a dock that lives on your desktop and shows what Claude/codex/Pi sessions are live, what PRs shipped, and where code conflicts are forming. its called Slashtalk - cut the chatter & work faster - slashtalk.com
English
michael s galpert retweetledi
The question you should be asking yourself is “when and how quickly is it gonna decelerate and why?”
Jesse🔸⏹️@PoliticalKiwi
At its current exponential growth, Anthropic's annualized revenue will hit 100% of global GDP in early 2028. Do I think this will happen? No. Is it insane that this is the current trajectory, and we should all be preparing for AI to rapidly change the world we live in? Yes.
English
i read about using plus.codes on here last night to be able to precisely direct a @Tesla where to park on FSD. it is an amazing unlock, worked great this morning. get the plus code from the site and just put it into navigation.
English
off to Shanghai for @clawcon’s first event in China
luma.com/clawconshanghai
English
michael s galpert retweetledi
Introducing the Printing Press, a CLI-factory and a CLI-library. Built with @trevin. 🏭🖨📚
Most APIs suck for agents. Most MCPs suck for agents. Most official CLIs suck for agents. They waste tokens and time. @steipete started making his own because of this.
📚 A Library of agent-native CLIs you install today (Linear, ESPN, Flight GOAT (Google Flights + Kayak nonstop), Contact Goat (LinkedIn + Happenstance + Deepline more) +30+ more)
🏭 A factory that prints new ones for any service - just type /printing-press
CLIs are fast, local, SQLite-backed. Work in Claude Code, Codex, OpenClaw, Hermes.
🌐 printingpress.dev
English
@nickgraynews @realGeorgeHotz It’s prob gonna double in next 3 years
English
After today’s $AMD earnings I’m reminded of @realGeorgeHotz’s $250k yolo play 14 months ago which is sitting at a cool $1M rn
geohot.github.io//blog/jekyll/u…
English
There's a topic going around that the prompt is the bottleneck now, not the model.
That’s true, probably. But for me, it's more complicated than that.
I rolled back from Opus 4.7 to 4.6 last week. Not because 4.7 is bad. But because my entire system — skills, memory layers, hooks, months of small corrections — was shaped around how 4.6 reads my intent.
It's like when your roommate "organizes" the kitchen. Everything's technically in a better spot. You still can't find the spatula at 7am.
When the model gets more literal, everything I've wired into it shifts. You don't notice it in one output. You notice it in twenty. Something's just... off. Like switching from iOS to Android and your thumb keeps reaching for a button that isn't there anymore.
The people adjusting their prompts are mostly adjusting production harnesses. Fixed instructions bolted onto an API. That's a solvable problem — you rewrite the system prompt, run your evals, ship it.
My situation is different. I'm in interactive mode. I'm in conversation with this thing every day. That relationship has texture to it. We have inside jokes (sort of). It knows when I'm being vague on purpose vs. when I genuinely don't know what I want yet.
4.7 stopped reading between my lines. Technically correct. Spiritually wrong.
"Prompt better" assumes the prompt is a string you type. For me it's ten months of accumulated context. Good luck migrating that in an afternoon.
Alex Prompter@alex_prompter
Both OpenAI and Anthropic just released official prompting guides. Both say the same thing. Your old prompts don’t work anymore. But for opposite reasons. Claude Opus 4.7 stopped guessing what you meant. It does exactly what you type. Nothing more, nothing less. Vague instructions that worked on 4.6? They now produce narrow, literal, sometimes worse results. Not because the model got dumber. Because it stopped compensating for sloppy thinking. GPT-5.5 went the other direction. OpenAI’s guide literally says: “Don’t carry over instructions from older prompt stacks.” Legacy prompts over-specify the process because older models needed hand-holding. GPT-5.5 doesn’t. That extra detail now creates noise and produces mechanical output. Claude got more literal. GPT got more autonomous. Both now punish the same thing: prompts written without clear thinking behind them. One developer on Reddit captured it perfectly after analyzing hundreds of community posts. The complaints tracked almost perfectly with prompt specificity. Precise prompts got better results on 4.7. Vague prompts got worse. The model didn’t regress. The prompts did. OpenAI’s new framework is “outcome-first prompting.” Describe what good looks like. Define success criteria. Set constraints. Then get out of the way. The model picks the path. Anthropic’s framework is the inverse: be surgically specific about what you want, because the model won’t fill in your blanks anymore. Two different architectures. Two different philosophies. One identical conclusion: the person writing the prompt is now the bottleneck, not the model. Boris Cherny, the engineer who built Claude Code, posted on launch day that even he needed a few days to adjust. That post got 936 likes. Meanwhile, Anthropic increased rate limits for all subscribers because the new tokenizer uses up to 35% more tokens on the same input. The model is more expensive to run lazily. Cheaper to run precisely. The models are converging in capability. The gap between good and bad output is no longer about which model you pick. It’s about the 2 minutes of structured thinking you do before you type anything. That thinking system is the skill. The prompt is just what it produces.
English