Agents and tooling

Coding agents and editors, agent SDKs and the protocols they speak, evaluation and observability, serving engines and local inference.

On the register

A2A after 1.0 watching
AGENTS.md as the shared instruction file watching
Agent Client Protocol in editors watching
Agent Skills as a cross-tool format watching
Agent frameworks consolidating watching
Agent identity in directories and OAuth watching
Agent payment and checkout protocols competing watching
Agents reviewing and approving pull requests watching
Assistant and coding plan ladders watching
Audit access to agent sessions watching
Bills of materials for models and datasets watching
Cloudflare's Accountable crawler designation watching
Coding tools changing owner watching
Content Credentials in capture and display watching
Gateways between agents and the model API watching
Google's coding agents consolidating watching
Hosted runtimes that run an agent's sessions watching
MCP SDK tiering watching
MCP's 2026-07-28 revision and its deprecation clock watching
Machine-readable AI usage preferences watching
Machine-readable marking of generated content watching
Official MCP extensions reaching hosts watching
Paying for crawled content watching
Per-conversation and per-action pricing for agents watching
Permission modes that let agents act without a prompt watching
Plans moving from seats to usage watching
Plugin marketplaces for coding agents watching
Search and retrieval layers sold to agents watching
Signed requests from agents on the open web watching
Tools choosing the model per task watching
Vendors retiring hosted evaluation platforms watching
Voice in coding agents watching
WebMCP as a browser standard watching
Weight storage and formats on Hugging Face watching
Worktree isolation in coding agents watching
llms.txt files on the open web watching

Everything filed here

2026-09-22 GitHub Copilot weekly release rolls out tiered auto model selection · Agents and tooling · 1 source