Living Knowledge Base

· product-design, agents

What Makes a Knowledge Base "Living"?

The metaphor implies:

  • Growth — accumulates over time without effort
  • Metabolism — processes new information automatically
  • Responsiveness — changes when the world changes
  • Circulation — information flows in/out without manual pumping

Counter-example: what's "dead"?

  • A snapshot that decays the moment it's written
  • Requires someone to remember to update it
  • Disconnected from source of truth
  • You visit it; it doesn't come to you

The core distinction:

Dead = point-in-time artifact Living = continuously synchronized with reality

Two Components of "Living"

  1. Automatic updates — the data stays fresh
  2. Contextual surfacing — it appears when relevant

Automatic updates are foundational. Contextual surfacing is downstream — you can only surface what's already current.

Trigger Points & Sources of Truth

The key question: where does truth live, and what triggers a sync?

Knowledge Type Source of Truth Trigger Point Decay Rate
People/roles LinkedIn Profile change, job update Months
Funding/stage Crunchbase, news Event (round announced) Quarters
Product/features Website, changelog, docs Release, page change Weeks
Strategy/priorities Earnings calls, blog, interviews Announcement Quarters
Tech stack Job postings, BuiltWith, GitHub New posting, repo activity Months
Relationship context Your emails, meetings, notes Every interaction Days

Two types of triggers:

  • Push — source emits a signal (news article, funding announcement)
  • Pull — poll on schedule or before an interaction

Polling heuristic: Poll intensity = interaction frequency × decay rate

  • Company you talk to weekly → poll daily
  • Company you talked to once 6 months ago → poll on-demand

Internal vs External Sources

Internal External
What Your interactions, your team's knowledge Public/world information
Examples Emails, meetings, notes, CRM, Slack LinkedIn, news, Crunchbase, their website, job postings
Control You own it You observe it
Signal quality High (you were there) Noisy (filtered through public channels)
Trigger model Event-driven (hookable) Poll or aggregator

Key insight: Internal sources are naturally event-driven. External sources require polling.

Your email client knows when you email someone. LinkedIn doesn't notify you when a contact changes jobs.

What each captures:

  • External = their public state ("they raised Series B")
  • Internal = your relationship context ("we had 3 calls, they care about X")

Both needed for a complete picture.

The Two Update Loops

A living knowledge base has two distinct update loops:

  1. Internal loop — event-driven, hooks into your workflow

    • Email sent → update relationship context
    • Meeting ended → capture notes
    • Real-time, high fidelity

  2. External loop — scheduled polling + event triggers from aggregators

    • News API surfaces funding announcement
    • LinkedIn profile change detected
    • Best-effort, variable latency

Related