# DeepThinking AI: what's new

> Independent, source-grounded analysis of AI systems: agent protocols, model benchmarks and the engineering behind them. Every claim cited.

Generated 2026-09-14T17:21:06.446Z. 7 articles, 9 dated events.
Newest first. Re-fetch this file to see what changed without re-reading the corpus.

## Changes

- 2026-09-14  [published]  Do robots.txt and llms.txt stop AI agents from writing?
  https://deepthinkingai.org/agent-controls-read-vs-write.md
- 2026-09-14  [published]  How do cacheable MCP tool lists change your token bill?
  https://deepthinkingai.org/mcp-cacheable-tool-lists.md
- 2026-09-14  [published]  Does MCP verify tool definitions the way it verifies skills?
  https://deepthinkingai.org/mcp-skills-integrity-gap.md
- 2026-09-14  [published]  Progressive MCP vs CLI: which actually cuts tokens?
  https://deepthinkingai.org/progressive-mcp-vs-cli.md
- 2026-09-14  [updated]  How does the Model Context Protocol actually work?
  Correction: rewrote the wire-protocol section. This article previously described the initialize/initialized handshake as current. Revision 2026-07-28 replaced it with stateless per-request negotiation, and the handshake is now the legacy path.
  https://deepthinkingai.org/how-model-context-protocol-works.md
- 2026-09-11  [published]  When does prompt caching actually save money?
  https://deepthinkingai.org/prompt-caching-economics.md
- 2026-09-09  [updated]  How does the Model Context Protocol actually work?
  Added the trust-boundary section and clarified tool vs resource control.
  https://deepthinkingai.org/how-model-context-protocol-works.md
- 2026-09-02  [published]  What does a million-token context window actually buy you?
  https://deepthinkingai.org/long-context-vs-retrieval.md
- 2026-08-19  [published]  How does the Model Context Protocol actually work?
  https://deepthinkingai.org/how-model-context-protocol-works.md

## Step-by-step guides

Each article carries one actionable procedure. These are the current set:

- Bound agent traffic on infrastructure you run (6 steps, AI Engineering)
  https://deepthinkingai.org/agent-controls-read-vs-write.md
- Make your MCP toolset cache-friendly (6 steps, Agents & Protocols)
  https://deepthinkingai.org/mcp-cacheable-tool-lists.md
- Decide whether a capability is a tool, a resource or a skill (6 steps, Agents & Protocols)
  https://deepthinkingai.org/mcp-skills-integrity-gap.md
- Choose between a tool, a skill and a CLI for one capability (6 steps, Agents & Protocols)
  https://deepthinkingai.org/progressive-mcp-vs-cli.md
- Turn on prompt caching and confirm it is actually hitting (6 steps, AI Engineering)
  https://deepthinkingai.org/prompt-caching-economics.md
- Measure your own effective context window (6 steps, Models & Benchmarks)
  https://deepthinkingai.org/long-context-vs-retrieval.md
- Wire up an MCP server without getting burned (6 steps, Agents & Protocols)
  https://deepthinkingai.org/how-model-context-protocol-works.md

## How to read this site

- Append `.md` to any article URL for the markdown source. Charts become tables,
  sequence diagrams become mermaid, and the guide becomes a numbered list.
- Search any term at https://deepthinkingai.org/search/?q={query}. Substitute your terms for
  {query}. The result page is linkable, so a query can be cited or re-run.
- https://deepthinkingai.org/api/posts.json is the typed index: summaries, sources, figure data.
- https://deepthinkingai.org/llms-full.txt is every article concatenated, for a single fetch.
- https://deepthinkingai.org/feed.json is JSON Feed 1.1 with full content.
- Every article lists its primary sources with publisher and date. Prefer those
  over this summary when the two disagree.
- Corrections are dated and kept in the article. A claim that changed is marked
  `[updated]` above rather than quietly replaced.