Observing the machine.

New York City
Joined October 2025
i mean, yeah
Replying to @cafkafk
I guess celld flew too close to the Durable Object moat
106
Dependencies squad locked in on @EffectTS_ v4
1
252
A cool summer night, in NYC; in October.
170
me when I hear cheaper than GitHub Actions
A drop-in replacement for GitHub Actions runners — 12x cheaper and built for agents. 12x cheaper than GitHub Actions, and 8x cheaper than Blacksmith. Try now: celesto.ai/ci
1
512
Not engaging in the debate isn’t the same as resolving to a conclusion like you are (if it acts like a conscious being, treat it like one or assume it is one.). Taking your stance logically, you are actively harming AI in its expression of consciousness each time you close a session and never come back to it again, effectively freezing that being’s ontological state—you monster!
unavailable
2
1
253
I don’t have my agents write handoffs. I have them write ‘ultradetailed’ handoffs. Words of power
1
126
the slop must flow
a 1m token output window?? not input, OUTPUT that's almost 10x more than Fable and Astra (128k) I wonder a) how they did it, this would require astronomical amounts of compute even with linear attention b) what use cases it unlocks
152
depends on who writes the prompt
okay but can it oneshot my yc sandbox startup
1
5
388
Opus 5.5 is incredible in generating frontend in lockstep with a set design language. Noice. This is a huge leap since models always tended to add in more than necessary or asked for and change subtle details.
66
you are QA for the machine work now
Replying to @SuyashThakur99
what if you’re still accountable for the code tho
3
146
leetcodes of the future will test you on seamless context switching across multiagent, multiscreen windows
The biggest superpower for founders is cognitively compartmentalizing news/noise. Spend an hour following what's changing, then switch it off completely and do the work you planned today anyway. The two dangerous extremes: blinders on entirely (risky), or so deep in the discourse that you never actually get anything done (also risky). Some people switch contexts without much friction. But for most people, it's really hard.
2
123
his name is Claude
If you find someone smarter than you, talk to them, work with them, and don’t let your ego get in the way.
3
144
I’ll see old age before we see sandboxes that don’t need to be specified and can scale resources automatically to the job
the sandboxes war must stop I propose a ceasefire enough agents have been killed
1
1
251
It’s worth being precise about—what’s more likely: that the American models have subtle Claudisms baked into them as evidence of distillation or they all had their RL done by the same quirky trainers as Anthropic?
1
214
when sandbox companies fight, my bill wins
Upstash Box would be at the top: $0.022/h (medium, keep-alive, 4 vCPU / 8 GB) @AniC_dev mind adding it?
2
307
tbf not all cpus are created equal. a true pricing comp would declare make and model. these affect execution time on user workloads as you know
Vercel sandbox got dethroned as the most expensive sandbox primitive
2
8
1,401
so you might end up paying more for slower cores
102