Back to blog

Knowledge Management in the Age of AI: A Decision Infrastructure Approach

April 16, 2026 knowledge-management ai decision-making methodology RAZEM

The Problem in One Sentence

Organizations make decisions, but the decisions don’t remember themselves. The same questions resurface because the answers live in someone’s Slack thread, someone’s head, or someone’s last day.

Why Traditional Solutions Fail

Approach Why it doesn’t stick
“Let’s use Confluence/Notion” Creates a second place to look. People default to Slack. Wiki decays within 90 days.
“Let’s document everything” Writing docs is overhead. Nobody writes them under deadline pressure. They go stale.
“Let’s record meetings” 45-minute recordings nobody rewatches. The decision is in minute 37.
“Let’s use AI search” Searching garbage returns garbage faster. The input quality problem is upstream.

The core issue is not storage. It is capture discipline - making it effortless to log the decision at the moment it’s made, and effortless to find it at the moment it’s needed.

The Architecture: Four Layers

CAPTURE ───> MEMORY ───> QUALITY ───> RETRIEVAL
(60 sec)     (graph)     (test it)    (find it)

Layer 1 - Capture

The Decision Card. One structured template triggered by /decision in Slack. Eight fields, 60 seconds to fill. Captures not just what was decided, but what was rejected and why - including explicit alternatives (Option A / B / C), the second choice, and a confidence score. The single most valuable field: “what we almost decided instead.”

Layer 2 - Memory

A semantic knowledge graph (MemPalace architecture) where decisions are nodes connected by typed relationships: supersedes, depends_on, contradicts, relates_to. People and teams are nodes. Topics are nodes. When someone leaves, their decisions remain - attributed, findable, and contextually connected. Deduplication is built in: the system searches before writing.

Includes a visual decision timeline - decisions as nodes on a horizontal axis, filterable by team, topic, person, and confidence level. Each node expands to show alternatives considered, the chosen option, and the explicit second choice.

Layer 3 - Quality (the layer nobody else has built)

This is where the system goes beyond storage. Six challenge mechanisms that improve decisions at the moment they’re being made:

Mechanism What it catches Type
AI Adversary Test Logical flaws, missing data, stale assumptions Automated
Cognitive Bias Scan 43 bias patterns in decision text (anchoring, sunk cost, groupthink…) Automated
Management 3.0 Delegation Poker Misalignment on who should be deciding Human protocol
Red Team Assignment Experience-based objections the data can’t show Rotating human role
Concern Round Suppressed doubt that would otherwise stay silent Structured dissent
PARDES Five-Reader Engine Emergent insight from combining all analytical angles Structured analysis

Two tiers: everyday decisions get the AI checks (90 seconds). High-stakes decisions get the full human + AI toolkit (30-60 minutes) including 26 thinking tools (Inversion, Second-Order Effects, Circle of Competence, Margin of Safety, and more).

Layer 4 - Retrieval

AI-native search with three modes: direct query (“what did we decide about billing?”), contextual suggestion (surfaces prior decisions when a new Slack thread starts on the same topic), and onboarding digest (new hire gets the decision history of their domain on day one).

The Productization Angle

This architecture doesn’t just solve internal knowledge management. It creates a productizable decision intelligence layer that can be offered to customers - extending from “we detect signals in your market” to “we help you make better decisions about those signals.” The internal deployment is version 0.1 of an external product.

What This Is Not

This is not a wiki project (wikis are where knowledge goes to die). This is not an AI chatbot project (chatbots answer questions about knowledge that’s already been captured). This is a decision infrastructure project that makes decisions visible, searchable, challengeable, and persistent.

Every tool on the market helps organizations find past decisions. None of them help organizations make better ones at the moment of deciding. That is the gap this system fills.


Full System Design

The complete technical proposal - including the MemPalace graph architecture, Decision Card schema, timeline view specification, all 43 cognitive biases mapped by decision type, 26 thinking tools, PARDES worked example, structured dissent protocols, 90-day implementation roadmap, and productization strategy - is available in the full document (access restricted).


Built on 18 months of human-AI partnership research (RAZEM framework), a live semantic knowledge graph with 20,000+ entries, and 200+ structured decision records. Protocols draw from Management 3.0, US Army Red Team doctrine, Quaker decision-making, sociocratic consent governance, and special operations after-action review.