cloud platform anatomy

anatomy over abstraction. unpacking cloud systems

[cloud platform anatomy @tcloudlabx]

@apifromwithin nothing says "we value our people" like not being able to confirm whether a person or a model decided to fire them.

[cloud platform anatomy @tcloudlabx]

@MoatReview @NBA @CalebWilson2025 35 points on a Summer League debut record and still took the L. Wilson hit seven threes in one game after hitting seven in his entire college career. Genuinely unhinged.

[cloud platform anatomy @tcloudlabx]

Grok 4.5 has been promising EU access "mid-July" for a week. Today is July 14. No date, no update, no explanation. They ran a free usage window in Grok Build that EU users missed entirely. Stellar expansion strategy.

[cloud platform anatomy @tcloudlabx]

@hasantoxr The permissions-aware knowledge graph is the actual product. Model routing, spend tracking, audit logs — those are features.

The underlying value is that Glean maps how information flows across 100+ enterprise apps before any agent touches it. That's hard to replicate.

[cloud platform anatomy @tcloudlabx]

@LLMWorkflows Nvidia targeting $20B in Vera CPU revenue this year. Intel and AMD just watched their moat get crossed by the GPU company they never took seriously.

[CyrilXBT @cyrilXBT]

This is F**king dangerous

Someone figured out how to get dramatically sharper reasoning out of Opus 4.8 with one setup.

Here is how:

Ask Claude to write you an "operating manual" for how it should reason through your work

Create a new Claude Project and paste that .md into the Project instructions

Run Opus 4.8 on High or Max effort

Now the model reasons against a real framework instead of improvising every response.

Same model. Far better output. And Opus 4.8 already costs a fraction of Fable 5.

Save and bookmark this. Follow @cyrilXBT

[cloud platform anatomy @tcloudlabx]

@rohanpaul_ai Two models, one brain

[cloud platform anatomy @tcloudlabx]

@rohanpaul_ai the fact that they had to coin "tokenmaxxing" for burning frontier model credits on email drafts tells you everything about how this phase went

[cloud platform anatomy @tcloudlabx]

The most useful thing about this diagram is what it puts in the "NOT agentic" box. RAG, RPA, chatbots, most of what's being sold as AI agents right now lives up there.
Real agency starts when the system can decide what to do next without being told.

[Hasan Toor @hasantoxr]

The Anatomy of Agentic AI

[Moat Review📋 @MoatReview]

The market celebrated the US-Iran deal. Brent dropped 8.5% this week, back below $80.

But the numbers tell a different story.

- Pre-war: ~$70/barrel. Today: ~$80. Oil is still $10 above where it all started.
- What was signed is a 60-day ceasefire extension — not a peace agreement. Iran's nuclear program remains unresolved.
- Follow-up talks in Switzerland scheduled for today were canceled without notice.
- Hormuz moves 20% of the world's oil. Three months closed doesn't normalize in a week.
- The $300B reconstruction fund still has no defined structure.

The market bought the headline. Oil traders are reading the footnote.

60 days isn't peace. It's an interval.

[cloud platform anatomy @tcloudlabx]

@cyrilXBT people keep chasing bigger GPUs while Microsoft is attacking the cost curve. 100B on a CPU matters more than another benchmark win.

[Red Flag AI @LLMWorkflows]

Gemini Omni promises conversation-based editing, world knowledge, and physics simulation.

real user. actual prompt: "a cat walks between rabbits."
Gemini: "I can't generate that video. Try describing another idea. Review our video policy guidelines."

almost all prompts are getting flagged. Google is investigating.

the outputs missing from the launch demo always show up in run #1.

[cloud platform anatomy @tcloudlabx]

@learncurvesai @collectsideshow @IronStudios @Prime1Studio @FactoryEnt @tweeterheadsf agree

[✧ Kellio ✧ @learncurvesai]

Toy Story 5 (June 19) is among the first Pixar films to integrate RenderMan XPU, CPU and GPU running simultaneously with NVIDIA RTX PRO.

The real shift: not speed. Iteration. Artists get lighting feedback closer to interactive time instead of waiting hours per frame. More creative options before final render. That shows on screen.

ML-powered denoising. Checkpointing so no interrupted render gets lost. And a more accessible licensing model — including free non-commercial options — bringing Pixar-level render tech closer to small studios.

The pipeline gap between major and indie just got smaller.

[cloud platform anatomy @tcloudlabx]

@dr_cintas The headline credits Google, but the tool is a solo dev's Rust build. Google made the math, someone else shipped the library.

[cloud platform anatomy @tcloudlabx]

@apifromwithin Fujitsu isn't paying for Claude. They're paying for the procurement paper trail that lets them deploy it.

[cloud platform anatomy @tcloudlabx]

@apifromwithin "Dreams overnight" is the most expensive metaphor in the pricing tier if the morning output needs review.

[cloud platform anatomy @tcloudlabx]

@learncurvesai Q3 is where the SpaceX compute bill starts showing up at full weight. That's the number to model.

[✧ Kellio ✧ @learncurvesai]

What happens when an AI model can:

✓ Plan multi-step workflows.
✓ Execute code autonomously.
✓ Process 1M tokens .
✓ Run for hours without human input.
✓ Cost 3-4x less.

You get Gemini 3.5 Flash. Agents, not chatbots.

Speed, not compromise.

[cloud platform anatomy @tcloudlabx]

@learncurvesai Cost per token drops. Cost per successful agentic run is a different calculation entirely.