Skip to main content

​
First-Class, Not Afterthoughts

In Memex AI, every non-obvious design choice is a Decision with five properties.

Stable ID

D1, D2, … referenceable from any work item, blueprint, or conversation.

Status

Open, Leaning, or Resolved.

Options

The plausible alternatives, with trade-offs.

Resolution

What was chosen, why, and what was rejected.
Plus impact links: which work items are blocked, which blueprints are affected.
Decisions are not documentation. They are active objects in the system. An open decision is a blocker. A resolved decision is a constraint. When priorities change, you re-open and re-resolve decisions, and the system traces the downstream impact automatically.

​
Why decisions are first-class

For AI agents

An agent asked to implement a caching layer checks Memex AI first. It sees D7: Cache invalidation strategy is still open with three options under consideration. The agent stops and reports this rather than guessing. The team resolves D7, and every agent working on related features immediately has the answer.

For humans

When a stakeholder asks “why did we build it this way?”, the answer isn’t buried in a Slack thread from four months ago. It’s D7, resolved on March 3rd, with the full rationale and rejected alternatives preserved.

​
Where decisions come from

Most decisions in a healthy Memex graph are not created by someone pausing to fill in a form. They’re captured passively from human–AI planning conversations and surfaced for lightweight review. See Decision Extraction for how that works.

​
Anatomy of a decision

D7 · Cache invalidation strategy                    [OPEN]
Strategy: S3 — Proactive Role Discovery

Options
  A. Time-based TTL (simple, may serve stale)
  B. Write-through invalidation (complex, always fresh)
  C. Event-driven via Pub/Sub (flexible, more moving parts)

Blocks
  WI-3  Agent conversation flow
  WI-4  Discovery integration

Affects blueprints
  caching, data-model
When resolved, the record keeps the options and rejected alternatives. Future contributors (human and AI) can see not just what was chosen but what was considered and why the others lost.