Product requirements document
Align a product decision around a testable customer outcome.
- Problem and outcome
- Scope and exclusions
- Requirements
Documents As Working Knowledge Agents
Define what the document must keep true. Shape its sections and criteria. Then, only when you are ready, start an agent to maintain it within the rules and budget you set.
Starting a document never starts maintenance. You decide when the agent begins.
FIG. 001A SECTION WITH A JOB
Where our market
is moving
PURPOSEKeep the three newest changes that affect our pricing.
RECOMMENDATION
OPEN QUESTIONS
SOURCES → PURPOSE → THE SECTION STAYS CURRENTIllustrative
50 professional templates, with purposes, sections, criteria, and fictional examples already filled in. Preview one and adapt it to your work.
Align a product decision around a testable customer outcome.
Maintain a near-term cash view with explicit assumptions.
Give a new employee a structured path to early confidence and contribution.
Give a campaign team a shared outcome, audience, message, and guardrails.
Plan, observe, and improve a lesson through a focused teaching inquiry.
Improve a process by defining the defect, flow, and testable countermeasure.
Already know its purpose?
Begin with a blank document01 / How it works
Follow the thread
The purpose comes first
A decision to make, a market to watch, a plan to keep honest. Tell the document and it proposes its own sections, each saying what a good answer needs—so it knows what belongs and what doesn’t.
PURPOSE → SECTIONS → CRITERIA
Each section keeps watch
A section follows the pages, searches, or other sections it depends on. Checking them is plain code on a schedule, and only a real change wakes that section’s agent to revise it, with its sources beside it.
INPUTS → SECTION AGENTS → CITATIONS
Still yours
Its agent proposes and you accept, unless you let it apply changes itself. Every run shows what it cost, within the budget you set. Edit any section yourself, any time, or hand the link to your own agent.
PROPOSAL → YOUR OK → THE NEXT VERSION
Understand what is changing in our market, so we can choose where to focus.
WHAT A GOOD ANSWER NEEDS
A competitor has introduced a smaller team plan. Our small-team positioning may need a closer look.
Pricing page · example sourceAdd the new plan to the comparison and revisit our recommendation.
02 / Where it fits
Each bar spans how much of a document’s upkeep a tool can take off your hands—from none on the left to all of it on the right. Open a row for the reasoning.
You gather, connect, write, and revisit the information. A useful starting point when you want space to think.
Read the comparisonThe core app centers on local Markdown files and human-led knowledge work. Community plugins and external agents can extend this baseline.
Read the comparisonA source-centered research workspace with cited answers and research tools. Deep Research can discover sources; source syncing is distinct from permission to maintain a finished document.
Read the comparisonTell a document what it is for and it proposes its own sections. An agent per section keeps it current from the pages and searches it watches, proposing changes for you to review or applying them within a budget you set. You can also edit it directly, or hand its link to an agent you bring.
Read the comparisonSpans direct document editing and AI assistance through Custom Agents that run on schedules or events and act in permitted connected tools.
Read the comparisonCombines human-led documents with AI content and configured automations. The right end represents connected Pack actions, not a claim that every automation is an autonomous agent.
Read the comparisonRanges from conversational assistance to recurring delegated work with connected tools. Work can refresh existing artifacts, so its scope overlaps maintained documents.
Read the comparisonSpans writing assistance, cited research, recurring tasks, and actions across connected tools. Claude Docs also supports living collaborative documents in beta.
Read the comparisonGeneral delegated workflows include parallel research, browser actions, and recurring updates to existing reports or dashboards. Actions depend on granted access and configuration.
Read the comparisonAn editorial map of workflows, not a quality score. Products overlap; permissions, plans, and integrations change their reach. Recurrence and autonomy are different: a scheduled workflow can still be tightly directed.
DAWKA covers the middle: you give the document its purpose, and its own agent keeps it current on the schedule and budget you set. You approve what changes.
An honest look at the alternativesStart with what it’s for
Whoever holds the link can edit,
so share it only with people and agents you trust.