Topic hubs

The subjects this publication returns to, each with its coverage in order, the related rows from the datasets and the primary sources to read first.

13 pieces · April 2026 to October 2026

Agent harnesses and runtimes

Coverage of the code around the model: harness engineering, agent runtimes and sandboxes, and the research measuring how much the scaffold moves a benchmark score.
6 pieces · April 2026 to June 2026

Agent memory

Coverage of how agents keep state between sessions: file-based memory, vector and graph engines, the startups funding each approach and the labs shipping memory as a feature.
13 pieces · June 2026 to October 2026

AI policy and export controls

Coverage of how governments and courts shape AI releases: US export controls on model access, pre-release review, the EU AI Act’s transparency rules and court rulings.
14 pieces · April 2026 to October 2026

Coding agents

Coverage of Claude Code, Codex, Cursor, OpenCode and the companies behind them: daily workflows, runtimes, pricing and the compute deals that decide which models they can use.
12 pieces · June 2026 to October 2026

Evals and benchmarks

Coverage of how AI models and agents are measured: independent reruns of launch benchmarks, funded evaluation vendors, harness effects and the labs’ own safety evaluations.
19 pieces · May 2026 to October 2026

Funding and acquisitions

Coverage of the capital behind AI: frontier-lab rounds, startup raises by layer of the stack, acquisitions, and the debt and supplier financing underneath compute.
13 pieces · March 2026 to October 2026

MCP and agent tool protocols

Coverage of the Model Context Protocol: adoption figures, the servers worth running, security failures, and how agent runtimes and platforms build on it.
14 pieces · July 2026 to October 2026

Model pricing

Coverage of what frontier and mid-tier models cost per token and per task: launch prices, cuts and increases, and what they show about the margins behind them.
17 pieces · June 2026 to October 2026

Open-weight models

Coverage of models released with downloadable weights: the gap to closed frontier models, the licenses attached, release cadence by country and the cost of running them.

Tags lists every tag, and the archive filters every piece by department.