Decision Latency & Personal Knowledge Systems

The bottleneck on smart decisions is no longer information - it’s the gap between capture and synthesis.

My take

The frame I use for my own work is decision latency: the time between a useful signal landing in my inbox and that signal actually changing how I think or act. That gap is where most knowledge work quietly fails. We have unprecedented access to information and unprecedented difficulty turning it into a position we can defend.

The naïve answer is “use AI to read more.” That makes the problem worse — it widens the funnel without improving the synthesis. The better answer is to wire AI into the synthesis layer: a personal knowledge system that ingests sources, integrates them into living concept pages, and surfaces contradictions when they appear. Build once, compound forever. That’s the system this site runs on.

What I’m watching: how attention management plays out when one person is orchestrating multiple concurrent agent sessions, and whether the next generation of PKM tools is built around durable concept graphs rather than another note-taking surface.

---

Everything above the divider is mine. Everything below is auto-assembled daily from my knowledge base — individual links and summaries may be stale or off-target. Last refreshed: 2026-05-10.

What’s shifted recently

• AI Attention Management (updated 2026-05-09)
  AI-induced attention deficit is a cognitive overload pattern that emerges when a single person orchestrates multiple concurrent AI agent sessions. — source · source · source

• Cross Agent Persistent Memory MCP (updated 2026-05-09)
  Cross-agent persistent memory via MCP is a local-first infrastructure pattern in which a single SQLite-backed store — exposed as an MCP server — gives every coding agent on a mach… — source · source · source

• Notebooklm As Research Substrate (updated 2026-05-09)
  NotebookLM (backed by Google Gemini) is a document-grounded AI workspace that accepts sources — PDFs, URLs, YouTube videos, articles — and enables chat, synthesis, and audio overv… — source · source · source

• Agent Memory Architecture (updated 2026-05-07)
  Agent memory architecture refers to the set of mechanisms by which AI coding agents and AI coworkers maintain context that persists beyond a single session, enabling continuity ac… — source · source · source

• AI Prompt As Curriculum (updated 2026-05-07)
  Prompt-as-curriculum is the practice of using structured AI prompts — rather than purchased courses or fixed institutional syllabi — as the primary mechanism for designing and del… — source · source · source

• LLM Hallucination Citation Verification Benchmark (updated 2026-05-05)
  A citation-verification benchmark jointly produced by EPFL and the Max Planck Institute uses 950 questions across legal, medical, research, and coding domains to measure how often… — source · source · source

• Agent Session Memory Loss Project Context (updated 2026-05-04)
  Coding agents and LLM-based assistants, including Claude Code, discard all conversational state when a session ends — there is no native mechanism to carry forward decisions made,… — source · source · source

The ideas I keep coming back to

Currently active (last 30 days):

• AI Attention Management — AI-induced attention deficit is a cognitive overload pattern that emerges when a single person orchestrates multiple concurrent AI agent sessions.
• Cross Agent Persistent Memory MCP — Cross-agent persistent memory via MCP is a local-first infrastructure pattern in which a single SQLite-backed store — exposed as an MCP server — gives every coding agent on a mach…
• Notebooklm As Research Substrate — NotebookLM (backed by Google Gemini) is a document-grounded AI workspace that accepts sources — PDFs, URLs, YouTube videos, articles — and enables chat, synthesis, and audio overv…
• Agent Memory Architecture — Agent memory architecture refers to the set of mechanisms by which AI coding agents and AI coworkers maintain context that persists beyond a single session, enabling continuity ac…
• AI Prompt As Curriculum — Prompt-as-curriculum is the practice of using structured AI prompts — rather than purchased courses or fixed institutional syllabi — as the primary mechanism for designing and del…
• LLM Hallucination Citation Verification Benchmark — A citation-verification benchmark jointly produced by EPFL and the Max Planck Institute uses 950 questions across legal, medical, research, and coding domains to measure how often…
• Agent Session Memory Loss Project Context — Coding agents and LLM-based assistants, including Claude Code, discard all conversational state when a session ends — there is no native mechanism to carry forward decisions made,…
• LLM Knowledge Management — LLM knowledge management is the practice of using language models as the primary agent for building, maintaining, and querying a personal knowledge base — typically structured as…

Who I’m watching

• Andrej Karpathy (person) — Andrej Karpathy is a researcher and educator who co-founded OpenAI and led Tesla’s Autopilot vision team.
• Andrew Ng (person) — Andrew Ng is a long-standing AI researcher and educator — co-founder of Coursera and Google Brain, founder of DeepLearning.AI, and a frequent commentator on practical AI deploymen…
• Garry Tan (person) — Garry Tan is the president and CEO of Y Combinator, and one of the most visible public commentators on AI coding tools, startup strategy, and AI security risk.
• Jensen Huang (person) — Jensen Huang is co-founder and CEO of NVIDIA, which under his leadership became the world’s most valuable company by capitalizing on the AI infrastructure buildout.

Sources I’ve been drawing on

• x.com — cited in AI Attention Management
• x.com — cited in AI Attention Management
• x.com — cited in AI Attention Management
• x.com — cited in AI Attention Management
• x.com — cited in AI Attention Management
• hindsight.vectorize.io — cited in Cross Agent Persistent Memory MCP
• github.com — cited in Cross Agent Persistent Memory MCP
• dev.to — cited in Cross Agent Persistent Memory MCP
• medium.com — cited in Cross Agent Persistent Memory MCP
• azarconsulting.com.au — cited in Cross Agent Persistent Memory MCP
• markanamedia.com — cited in Cross Agent Persistent Memory MCP
• www.decodingai.com — cited in Cross Agent Persistent Memory MCP