AI Innovation Lead @ Zapier · Optimist · Investor · Author of the buildermanifesto.com -- My views!
New York, NY
- Tweets 12,824
- Following 607
- Followers 1,529
- Likes 19,785
If you block your website with a signup/subscription wall.
I will simply leave, don't even bother to sign up.
Same rule applies to agents. Humans and agents seek the path of least resistance.
46
End of an era. Start of another.
Warren Buffett is stepping down as chairman of Berkshire Hathaway $BRK.B, effective immediately - CNBC
61
Now imagine this, in a real life AR video game with on device inference, almost zero latency.
TypeGPU + ruNNtime + Jev @typesafeai is a very fun combo :D
ruNNtime gives me efficient local inference, TypeGPU lets inference and rendering share GPU resources directly with zero copy. That’s 3 separate NN inferences plus rendering, all happening in realtime
Since we control the pipeline, Jev can just sit in the middle and add the semantic bit.
camera + mic → Moonshine + YOLO26 + DepthART → Jev → lights, shadows and bloom
1
124
FINALLY!
Next: Chat with Claude Code!
Claude Cowork and chat are merging into one Claude.
Ask a quick question or hand over a report, and Claude takes it from there, even after you close your laptop. If something's unclear, Claude asks—you keep the final say.
Rolling out to Pro and Max over the next few weeks.
1
129
Jokes apart, so many good games and experiences will come out of this.
Also I call it now, this is the killer of the iPad mini.
1,279
First time in Europe after 6 months in the US, impressions:
- I've forgotten they use paper straws
- They charge plastic fee on a Starbucks coffee
- Europemaxxing and aging population is real, friends of same age think like my parents
- Food is so good 🤤
I will report more
2
3
169
Astra made me go back to projects that I never started either because of time or because of cost, and make me wanna build them.
Even one-shot them.
Imagine the scale of this to much bigger problems... Security, science, health, finance.
94
Last Friday, I wanted to control my home Mac through @Astropad. The local server had an issue and I couldn't connect
I asked Grok @bot to restart it and it worked.
Similarly you can hand-off any kind of task to your local compute, remotely.
Personal Mac fleets are apparently a thing now and nobody told the interface designers.
The replies to this are wild. People have Minis, Studios and old MacBooks scattered around doing different jobs, connected with Tailscale, SSH and various remote desktop tools.
If you’re one of these people, please add your cursed setup to the thread. I’m collecting them.
1
6
1,773
Hooks are amazing, for this job.
A Spotify engineer built a Claude Code plugin that blocks big file reads at the hook layer and reroutes them to a cheap Gemini worker instead of Claude. His first version was routing advice in CLAUDE.md. Claude ignored it. The hook doesn't ask, it blocks. He measured a 90% drop in token spend on one Java monorepo.
151
"...most of what people use frontier models for in a coding workflow..."
This happens due to the state of UX, you can't switch between models mid-chat, so you either solve it with a custom router or depend on provider's auto-mode, that's not always tailored to your needs.
Spotify's engineering team cut Claude Code token usage by 90% with a two-model routing approach they call Portal. The mechanism is worth understanding because it's not prompt engineering, it's hard blocks at the architecture level.
Two cheaper "assistant" models handle the tasks that don't need the expensive model: opening files, writing repetitive code, boilerplate. The expensive model only touches novel reasoning. Files over 350 lines get blocked from the expensive model entirely, not suggested away from it, blocked.
They tried rules-as-instructions first. It didn't hold. Engineers (and the model) found ways around soft rules. Hard blocks at the routing layer worked where prompt-level guidance failed.
90% token reduction is a big number. The implication is that most of what people use frontier models for in a coding workflow doesn't actually require frontier-level reasoning. Routing to capability level rather than defaulting to the best model available is a legitimate architecture pattern.
Article is worth reading in full.
nitter.net/undefinedKi/status/209…
139
Astra's performance leap is similar to Opus 4.5.
The moment you set a /goal, you can feel how eagerly but concisely tries to solve a problem, without redundant tool calls or misses.
Closer or not to AGI the productivity from now will change forever.
71
