Card graphs, spec drift, decision logs, persistent agent memory — and what actually changes when the substrate underneath your AI agents is something other than chat history.
Obsidian and Notion sit at opposite ends of a real design spectrum: local-first Markdown files vs cloud-based blocks, personal PKM vs team collaboration. A no-spin, third-party comparison across seven dimensions — plus a clear framework for which one you should actually pick.
Product managers own what gets built and why — not the building itself. A clear guide to the core responsibilities, the real day-to-day, the skills that matter, how the role differs from project manager and product owner, and the invisible work nobody warns you about.
A practical guide to customer feedback analysis — collect across channels, tag and quantify by frequency, impact, and segment, surface the real themes, and turn them into decisions. Plus the part most guides skip: why the line from insight to decision to shipped product keeps breaking.
SWOT, Porter's Five Forces, feature matrix, perceptual map, competitive matrix — what each is good at, when to use it, and a worked example for each. Plus the problem none of them solve: a competitor analysis is right the day you finish it and wrong the moment a rival reprices.
A PRD (product requirements document) describes what you're building and why, so the whole team can decide without re-asking you. What it answers, the parts it's made of, who writes it, how it differs from a spec, brief, or BRD — and the failure mode nobody warns you about.
Prompt engineering optimized the wording of one call. But agents run for many steps, and what decides their quality is what they're looking at when they read each instruction. That's context engineering — and why bigger context windows don't solve it.
A bigger context window raises the ceiling on what you can paste — but it isn't memory, and past a point it stops helping. Why a full window still forgets, and what actually keeps an AI consistent on long projects.
A practical guide to writing a PRD — the eight sections that earn their place, and the part everyone skips: why PRDs fail by drifting, not by being badly written, and what fixes that.
Chat history is the wrong shape for product thinking. Card graphs are a different substrate — one whose primary user is an AI agent collaborating with a human, not a human reading the document later.
Spec drift isn't a discipline problem. It's a substrate problem. Here's how dependency tracking solves it the same way invalidated build artifacts get solved — automatically.
A concrete workflow walkthrough. Brainstorm in Draftlize, implement in Claude Code via /draftlize:cite, mark built, watch for drift_detected when the spec moves.
ADRs have been around since 2011 — and have always atrophied. What changes when the audience isn't only humans? The AI agent becomes the discipline-enforcing reader.
Five PRD templates built on cards instead of prose. The structure makes drift visible the moment dependencies move. Each ships as a starter pack in Draftlize.
Why ChatGPT Memory / Claude Projects don't fully fix the re-explanation problem, and what actually does. Plus a practical 4-step switch guide.
The personal version. The bet behind the substrate, the secondary bets we got wrong (free tier, voice input, audience), and what dogfood taught us about who actually loves it.
The conceptual heart of Draftlize is a loop: chat with the agent, agent writes cards, cards become artifacts, artifacts feed future chats. 5 minutes to understand.
Why Draftlize shipped MCP server support on day one of the protocol. What the integration enables. 60-second setup with Claude Code, Cursor, or Claude Desktop.
Real cards, real edits, real cascade. A walkthrough of what Draftlize does when you change a single decision in a 47-card project — and the 12 dependents that flag automatically.
Obsidian / Roam / Notion are wonderful for human knowledge work. They're the wrong shape for AI context. The mismatch is structural — and explains why a new tooling category is emerging.