Unboxing the AI Stack

The stack

MetaPrompt files every post on one of ten layers, from agents and models at the top, through benchmarks, embodied AI and token economics, down to chips, data centers and the policy beneath them. Each layer keeps its color across the site.

L01

Agents

AI agents by the numbers: agent products and releases, agent frameworks, protocols such as MCP and A2A, computer-use and browser agents, and the payment rails agents use.

No posts yet
L02

Language Models

Large language models by the numbers: specs, context windows, API prices and rate limits, releases, open and closed weights, and dated leaderboard snapshots.

No posts yet
L03

Voice & Audio

Voice AI by the numbers: text-to-speech, speech-to-text and speech-to-speech APIs, consented voice cloning, realtime voice-agent latency and music generation.

No posts yet
L04

Image, Video & Avatars

Image, video and avatar generation by the numbers: the models and their APIs, talking-head and avatar APIs, and lip-sync.

No posts yet
L05

Benchmarks & AGI

How AI benchmarks work, dated cross-model leaderboard snapshots, contamination and saturation, and measured progress toward AGI with every forecast attributed.

No posts yet
L06

Embodied AI

The tech of embodied AI: robot foundation models and world models, humanoid and robot specs, actuators and onboard compute.

No posts yet
L07

Tokens & Economics

The economics of AI by the numbers: token demand and usage shares, price-per-token trends, inference cost, and AI capex and revenue as filed or reported.

No posts yet
L08

Chips & Memory

AI chips and memory by the numbers: GPUs, TPUs and custom accelerators, HBM and DRAM supply and prices, foundry and advanced packaging, and MLPerf results.

No posts yet
L09

Data Centers & Energy

AI data centers by the numbers: campuses and GPU clusters, power and cooling, capacity split into operational, under construction and announced, and grid demand.

No posts yet
L10

Policy & Standards

The rules around the AI stack: export controls, AI regulation such as the EU AI Act, safety frameworks and evaluations, and technical standards.

No posts yet