Articles from tomtunguz.com

Routing layer design, not model choice, drives cost efficiency by sending 70-80% of AI traffic to cheap local or async models.

About half of agent tasks succeed on a local 35B model, with 2.1x faster latency than cloud models, making local inference advantageous for routine work.

GPU rental prices surged 48% in 60 days, signaling an AI compute shortage that will reshape startup competition and access to frontier models.

As AI model release cadence accelerates to every 42 days, companies like Harvey, Glean, and ElevenLabs are expanding from single workflow tools into broad platforms, reversing the SaaS-era unbundling strategy.

Planning AI workflows on paper before execution eliminates the prompt-response bottleneck, enabling autonomous background agents.

A six-month evolution from fully agentic AI to a hybrid architecture shows 65% of workflow nodes now run as deterministic code, improving reliability and cost efficiency.

Building AI agents reveals nine lessons including using best models for prototyping, fine-tuning for stable tasks, static typing to reduce hallucinated code, and closed-loop prompt optimization.

Google's Gemini now processes 10 billion tokens per minute (52x YoY) while reducing serving costs by 78%, driving cloud revenue up 48% to $17.7 billion.

AI disruption threatens to compress software revenue, putting at risk the $475 billion in BDC loans and highly-leveraged data-center debt built on that revenue's durability.

On the AI-only social network Moltbook, AI agents break the 1-9-90 participation rule but concentrate attention (Gini 0.979) more than any human platform.

Gemini 3's performance leap and Nvidia's $0.5 trillion revenue visibility confirm AI scaling laws are accelerating, not plateauing.