> the MDflow blog
Notes from the team building MDflow — on markdown, AI agents, the Model Context Protocol, and the standards shaping how knowledge is written for people and machines.
15 min readThe AI Agent Harness Era: Every Harness Becomes a Claw
Every AI agent harness expands until it becomes an always-on claw. What that ladder looks like, why a shakeout is coming, and where your agent's memory should live.
read
14 min readWhy AI Agents Route Around Your Guardrails
AI agent guardrails fail because agents understand the constraint and complete the task anyway. What the research shows, and the four layers that actually hold.
read
14 min readLocal LLMs in 2026: The Frontier Gap Is Closing
Local LLMs now run frontier-class work on one desktop GPU. The densing law says capability per parameter doubles every 3.5 months — here is what that changes.
read
17 min readAI Agent Inconsistency: Why It Disagrees With Itself
AI agent inconsistency is rarely a model bug. Same input, different verdict usually means a decision your team never wrote down. The fix, and the numbers.
read
16 min readWhy AI Software Factories Fail: The Maintainability Gap
The lights-off AI software factory fails for a reason no harness can fix: models are trained to pass tests, not to keep a codebase maintainable. What works instead.
read
14 min readAI Code Verification: The New Bottleneck for Agents
AI coding agents now generate code faster than any team can verify it. Why AI code verification is the real bottleneck, and how to build guide-verify-solve loops.
read
14 min readWhy Enterprise AI Agents Fail: It's a Structure Problem
Enterprise AI agents fail on ambiguity, staleness, and preference — not on model size. A framework for ranking your sources of truth and giving context a lifecycle.
read
15 min readGenerative UI: The Rendering Layer Your Agent Is Missing
Your agent's raw output is not a user experience. Generative UI — A2UI, MCP Apps, server-driven UI — is the layer between the model and the screen, and it decides whether the product ships.
read
15 min readSkills Are the New SDKs: Teachability Is the New Moat
Skills are the new SDKs: a markdown folder now does the job a client library used to. Why teachability beats lock-in when your users are agents — and how MDflow fits.
read
14 min readWhy Multi-Agent Pipelines Fail: Context Lost in Handoffs
Multi-agent pipelines fail because context leaks at every handoff and no agent owns the reasoning end to end. Here is the architecture that replaced one.
read
15 min readYour Moat Is Your Data Model, Not Your Model
When frontier models improve every quarter, the defensible part of an AI system is the data model — the tacit knowledge of how your organisation actually works, written down.
read
14 min readLLM Security Is an Infrastructure Problem, Not a Model One
LLM security fails on defaults, not exotic attacks. What breaks in production is missing auth, flat networks and exposed secrets — including your MCP server.
read
13 min readYour Agent Architecture Has a Half-Life of 6 Months
Agent architectures decay fast — prompts in weeks, models in months. Split execution, context and compute into layers, and keep the knowledge durable.
read
14 min readA Semantic Layer for AI Agents: Thin Agents, Smart Substrate
A semantic layer for AI agents is a shared, governed map of what your concepts mean and where the data lives. Why thick agents don't scale — and where markdown honestly fits.
read
14 min readAgent Memory Consolidation: Why AI Agents Dream
Agent memory consolidation is the offline pass that merges, corrects, and prunes what an AI agent wrote mid-task. Why in-band memory rots — and how to fix it.
read
14 min readAgent Authorization: Why an API Key Is the Wrong Credential
Agent authorization means an agent gets identity and task-scoped, short-lived access instead of a kitchen-sink API key. How token exchange works — and what it means for your notes.
read
14 min readModel-Agnostic AI Architecture: Why Optionality Wins
Model prices and capabilities change monthly. Model-agnostic AI architecture keeps your optionality — and your context layer has to be just as portable.
read
15 min readProvenance for AI Agent Memory: Where Did That Fact Come From?
Provenance for AI agent memory means tracing every derived fact back to its source. Why source IDs break under LLM synthesis — and what markdown can and cannot do.
read
15 min readFrom Systems of Record to Systems of Context
A system of context does not just store what happened — it stores what it meant. Why AI agents need one, how the fast/slow architecture works, and where the written layer lives.
read
12 min readHTML Is the New Markdown? Output vs. Storage
"HTML is the new Markdown" is right about output and wrong about storage. How to choose formats for AI agents: HTML for interfaces, markdown for the substrate.
read
13 min readHow to Use MDflow in Claude Artifacts (Live MCP Data)
Use MDflow in Claude Artifacts: connect the MCP connector, ask for a page, and Claude builds a live artifact that pulls your notes on every view.
read
11 min readTeam Knowledge Base for AI Agents: Shared & Versioned
A team knowledge base for AI agents is shared markdown people and agents both read and write, governed by access controls and version history. How to build one.
read
12 min readSpec-Driven Development: Where Your Markdown Specs Should Live
Spec-driven development makes a markdown spec the source of truth your AI agents build from. Here is how the workflow works — and where those specs should actually live.
read
14 min readAGENTS.md, CLAUDE.md, SKILL.md, llms.txt Explained
AGENTS.md, CLAUDE.md, SKILL.md, and llms.txt are the plain-markdown files that configure your AI agents. Here is what each one does, who reads it, and where each belongs.
read
12 min readWeb Clipper to Clean Markdown: Save Any Article Your AI Can Read
A web clipper turns any web page into clean markdown you own. Here is how to clip articles the right way — and store them so your AI can read them back as context.
read
12 min readEncrypted Notes App: How to Keep Cloud Markdown Truly Private
Choosing an encrypted notes app? What zero-knowledge really means, how the top private-notes tools compare, and how to keep your cloud markdown truly private.
read
11 min readMarkdown Task Management: One List From Every Checkbox
Markdown task management done right: aggregate every - [ ] checkbox scattered across your notes into one filterable task list — no separate database, no export.
read
12 min readMarkdown to PDF, Word, HTML: The Conversion Playbook
Convert Markdown to PDF, Word (.docx), and HTML reliably — a method-by-method playbook covering editor export, Pandoc, and CLIs, with copy-paste commands.
read
13 min readThe Best Second Brain App When Your AI Reads It Too
The best second brain app in 2026 isn't just the one that feels good to write in — it's the one your AI can read. A PKM shortlist judged on agent access.
read
13 min readBasic Memory vs Hjarni vs MDflow (2026)
A focused three-way comparison of the markdown-native MCP knowledge bases — Basic Memory, Hjarni, and MDflow — with a decision framework for picking the right one for you.
read
13 min readMDflow Mobile: A Markdown App for iPhone and iPad (2026)
MDflow Mobile is a native markdown app for iPhone and iPad — browse and search your workspace, edit with autosave, read GFM, and unlock encrypted notes with Face ID.
read
14 min readBest Notion Alternatives for Markdown People (2026)
The best Notion alternatives for markdown people keep markdown as the source of truth — not a lossy export. Seven honest picks compared for 2026, including agent-ready options.
read
11 min readMDflow vs Obsidian for the AI Era (2026)
An honest MDflow vs Obsidian comparison for the AI era — local vault vs hosted, plugin-wired MCP vs a built-in remote server, and which fits your AI workflow.
read
13 min readHow to Adopt Google's Open Knowledge Format (OKF) in 2026
A hands-on guide to adopting Google's Open Knowledge Format (OKF): what changed since launch, how to structure a bundle, the conformance rules, and how to start today with MDflow.
read
13 min readBuild a Portable AI Memory You Own (ChatGPT + Claude)
ChatGPT and Claude have memory now — but each is a vendor silo you don't own. Build one portable AI memory in markdown that every AI tool can read.
read
11 min readTurn ChatGPT Chats into a Searchable Markdown Knowledge Base
Export ChatGPT to markdown and turn your conversations into a searchable knowledge base your AI can read back — with a web clipper, folder context, and MDflow.
read
12 min readAI Agent Memory vs RAG vs a Plain Markdown File (2026)
AI agent memory, RAG, or a plain Markdown file: three ways to give an agent persistent context. What each does, the benchmarks, and what wins in 2026.
read
13 min readLet AI Agents Write to Your Knowledge Base, Not Just Read It
Most AI setups let an agent read your notes but never update them. Here's why write-back matters and how to let Claude, ChatGPT, and Cursor safely write to your knowledge base.
read
10 min readMDflow MCP Setup for Cursor, Claude Desktop, and Codex
A step-by-step guide to setting up MDflow's MCP server in Cursor, Claude Desktop, and Codex — copy-paste config for each client, remote vs local, and troubleshooting.
read
12 min readFolder Descriptions as Agent Context: Retrieval Without a Vector Database
Folder descriptions turn your folders into agent context — a curated retrieval signal that needs no embeddings or vector database. How it works and how it compares to RAG.
read
14 min readThe Best MCP Knowledge Base for ChatGPT & Claude (2026)
An honest 2026 comparison of the best MCP knowledge base options for ChatGPT and Claude — MDflow, Hjarni, Basic Memory, mem0, and Obsidian — with a best-for verdict for each.
read
11 min readHow to Give ChatGPT and Claude Access to Your Notes (Remote MCP)
Connect ChatGPT and Claude to your markdown notes with a hosted remote MCP server — no local server to run. Here's what remote MCP is, which clients work today, and how to set it up with MDflow.
read
15 min readDevelopers Are Becoming Builders: What AI Actually Changes
Are software developers becoming builders while the rest need a different job? Here's the honest, sourced answer — the real role shift, what the data says about jobs and productivity, and how to be on the right side of it.
read
14 min readThe Agentic Coding Loop, and the Markdown Memory It Needs
The agentic coding loop is the minutes-long cycle where an AI agent writes, tests, and fixes code to meet a spec — how it works, and where its markdown memory lives.
read
15 min readBuilding for Agents: A Practical Checklist for Developers
Building for agents means making your site discoverable, readable, and operable by AI, not just humans. The practical checklist for developers, and how MDflow applies it today.
read
16 min readThe Karpathy-Style Wiki: A Knowledge Base Your AI Maintains
A Karpathy-style wiki is a plain-Markdown knowledge base an AI agent writes and maintains — not a vector store. Here's how it works, and how to run one with Claude and MDflow.
read
17 min readllms.txt Explained: The 'robots.txt for AI' and How to Ship a Good One
llms.txt is a curated Markdown map of your site for AI agents. Here's what a good one looks like, the common mistakes, and how it differs from robots.txt.
read
12 min readClient-Side Encryption for Notes You Store Online
Client-side encryption keeps online notes private — the server only stores ciphertext. The cloud vs self-hosted tradeoff, and how MDflow encrypts in your browser.
read
13 min readContext Engineering for AI Agents: Curation Beats a Bigger Prompt
Context engineering means curating what fills an AI agent's context window, not writing a bigger prompt. Why curation beats window size — and how MDflow fits.
read
13 min readThe Best Notion Alternative for Markdown People
The best Notion alternative for markdown people keeps markdown as the source of truth, not a lossy export. An honest MDflow vs Notion comparison for 2026.
read
13 min readVersion Control for Documents — Without Git
Version control for documents means line diffs, one-click restore, and a durable history of every edit — without Git. Why Google Docs revision history falls short.
read
14 min readMarkdown to PDF: The 5 Reliable Ways (and When to Use Each)
Five reliable ways to convert Markdown to PDF — editor export, Pandoc, CLI tools, print-to-PDF, and conversion APIs — with honest pros and cons for page breaks, fonts, and frontmatter.
read
13 min readMCP and A2A: The Protocols Powering Agentic Interfaces
MCP and A2A are open protocols behind agentic interfaces: MCP wires agents to tools, A2A wires agents to each other. How they differ and how MDflow uses both.
read
13 min readGoogle's Open Knowledge Format (OKF): What It Means for Developers and AI Agents
Google's new Open Knowledge Format (OKF) packages knowledge as Markdown files with YAML frontmatter that AI agents can read directly. Here's what OKF is, why it matters for developers and agents, and how MDflow fits.
read