# /init.news

> Practical AI automation, business growth, and discovery guides for owners wearing too many hats. Clear decisions, useful tools, and source-checked methods from /init.

Source-checked public guides for owners, builders and creative studios. Each Markdown note includes its canonical HTML address, publication and revision dates, AI-assisted authorship and inspected sources. Experiments are labelled; unknown results remain unknown.

## Start here
- [Editorial method](https://init.news/about.md): How sources, revisions and AI assistance are disclosed.
- [Content index](https://init.news/index.json): Structured answers, questions, dates, source links and related guides.
- [Complete public text](https://init.news/llms-full.txt): Full current articles with evidence and limitations.
- [Formats and reading routes](https://init.news/discovery/): Human-readable guide to the public formats.

## Agent systems
- [Before you add another agent, find the bottleneck.](https://init.news/notes/agent-systems-brief-2026-10-16.md): Five changes worth checking before you buy more agent capacity: stage-specific delegation, bounded shared memory, associative retrieval, MCP placement, and local tool-history fixes. Updated 2026-10-16.
- [A handoff is not done when the message is sent.](https://init.news/notes/agent-handoff-receipts.md): Record what the next worker received, which state it used, and what actually committed. A small receipt contract for durable agent workflows. Updated 2026-10-09.
- [Your agent remembered. Did it change the next action?](https://init.news/notes/agent-systems-brief-2026-10-09.md): Five source-checked changes in orchestration, execution evidence, memory validity, MCP discovery, and local inference that affect implementation choices. Updated 2026-10-09.
- [Before you buy another Mac, measure the whole request.](https://init.news/notes/apple-silicon-inference-evaluation.md): An evaluation protocol for MLX, Ollama, LM Studio, and exo that keeps useful output, waiting time, context, and memory in the same decision. Updated 2026-10-09.
- [The instructions still load. The advice may be wrong.](https://init.news/notes/engineering-claim-freshness-ledger.md): Before adding AI answers, give the instructions behind them an owner, a review trigger, and a correction path. A ten-page worksheet for busy teams. Updated 2026-10-20.
- [Your agent does not need every tool in the room.](https://init.news/notes/mcp-tool-discovery-context-budget.md): Separate the tool catalog from the model's working set. A practical design for discovery, schema loading, authorization, and context-budget tests. Updated 2026-10-09.
- [Your agent remembers the answer. Who tells it the answer changed?](https://init.news/notes/memory-provenance-invalidation.md): A small memory contract for source versions, valid time, dependency checks, and corrections that reach the answers built on an old fact. Updated 2026-10-09.
- [Before you add a graph, find the missing answer.](https://init.news/notes/graphrag-evaluation-before-migration.md): A decision worksheet for teams choosing between a simpler retrieval baseline and a graph-based memory system. Updated 2026-10-09.

## Optional
- [RSS](https://init.news/feed.xml): Publication and revision feed.
- [Sitemap](https://init.news/sitemap.xml): Canonical public HTML pages.

This reading map provides navigation. It does not establish indexing, citations or a ranking benefit.