Tech & macro: understand before you decide
Understand where we are before deciding where to go. This cluster translates technical shifts into decision consequences, without requiring an engineering background to follow.
It opens with a primer — 70 years of AI history in one article — then tracks the shifts that matter: SLMs set to replace 70 % of LLM calls inside agents at a fraction of the cost, the architectures taking shape after LLMs (CALM, world models) just as Yann LeCun leaves Meta, and the move of compute to the edge that renders part of the cloud obsolete.
It also follows the economics: the collapsing cost of software, DeepSeek pulling off the cheaper-transformer trick again, the DORA 2025 report on AI-assisted development, and interfaces disappearing — Zero UI and agentic chains, which raise a question of observability before they raise one of technology.
The cluster's pillar
All articles in this section
Beyond LLMs: CALM, World Models and the Future of AI
Yann LeCun leaves Meta for world models. 'LLMs are useful, not intelligent'. CALM 4× faster. Why your AI architecture must anticipate the post-LLM era.
DORA 2025 Report: AI Transforms Software Development
90% of devs use AI. +80% perceived productivity but +7.2% instability. Only 5% truly integrate. Guide to structural integration.
DeepSeek Strikes Again: More Stable, Cheaper Transformers
1 year ago: $600B Nvidia evaporated. Jan 1, 2026: new mHC paper. Token prices /1000 in 3 years. Does your AI roadmap survive a new 90% drop?
What If Elon Musk Uberized NVIDIA? 100M Teslas = 1000× Top Supercomputer
Amazon +$150B, Microsoft +$100B in data centers. 100M Teslas idle 95% of the time = 100 GW of compute. Uberization applied to AI compute.
NVIDIA Makes Your Cloud Architecture Obsolete
CES 2026: Jensen Huang ends the all-cloud era. Inference cost /10 with Rubin. AMD and Lenovo push Local First. AI leaves the screens.
OpenAI Is Netscape's Story: It Ends Badly
1995: Netscape revolutionizes the web, killed 3 years later by free IE. OpenAI lives on subscriptions, Google can offer free. Modularize your stack.
End of Human OS: Welcome to AI OS
Flowith OS: 1st OS designed for AI agents, not for you. Beta — agents winning at poker. Is your offering buyable by an AI without a screen?
SLMs Will Kill LLMs (And Nobody Talks About It in ExComs)
Phi-3 Mini: 3.8B params, 73% cheaper, 15× faster than 10× larger LLMs. NVIDIA: 70% of LLM calls replaceable by SLMs in autonomous agents.
Software Cost Is Collapsing: Code Like SMS in 2000
Replit, v0, Figma Make: building an app as easy as posting on TikTok. Content becomes app. RAG and CAG: the bridge to augmented decision-making.
Other sections
Does one of these hit home?
AI transformation, SAFe coaching or simply an outside opinion: 30 minutes is often enough to see clearly.
Book 30 min