Why You Need an AI Framework Before Your Staff Start Using AI Tools
AI Governance • IT Strategy • Software Engineering
Strategic Summary: Deploying AI models without a central operational framework introduces subtle inconsistency and session drift[cite: 1]. Geordie Hogarth outlines how Si Futures established centralized skill repositories, real-time event logging, and core architectural rules across its internal Meridian stack[cite: 1].
What Inconsistency in AI Tools Actually Looks Like
Nothing failed dramatically[cite: 1]. The problems were quieter and more subtle, which made them dangerous[cite: 1]:
- Session Documentation Drift: Different AI instances recorded progress inconsistently, generating sprawling notes that confused team members[cite: 1].
- Deferred Status Logging: Project updates were recorded at the end of long sessions rather than in real time, causing context loss[cite: 1].
- Isolated Knowledge: Lessons learned by one engineer’s AI session remained trapped in local history instead of propagating to the team[cite: 1].
- Infinite Session Compaction: Engineers frequently maintained continuous sessions for weeks, using “resume and compact” loops[cite: 1]. This bypassed initialization routines designed to execute at session start[cite: 1].
That final discovery reshaped our governance strategy: you cannot build guardrails around how you assume people work—you must design them around actual operational behavior[cite: 1].
Three Pillars of AI Governance
1. Automated Repository Sync: Centralized AI skills update automatically on session boot rather than relying on manual pulls[cite: 1].
2. Real-Time Event Logging: AIs update task records immediately upon completion, neutralizing infinite session drift[cite: 1].
3. Inter-Tool Feedback Bus: AI agents flag rule friction directly to a central repository for continuous refinement[cite: 1].
Roads and Cobblestones: How Much to Standardise
The most frequent question I receive from business leaders is how tightly to enforce AI usage standards[cite: 1]. My answer is built on the concept of municipal infrastructure[cite: 1]:
A nation requires standardized highways; if every province built roads using different lane widths and sign rules, commerce would collapse[cite: 1]. But within a local village, cobblestone streets make perfect sense[cite: 1].
Enterprise AI frameworks function identically[cite: 1]. Across our entire framework, we maintain approximately twenty global rules[cite: 1]. Only five or six are non-negotiable hard requirements: API interfaces, database schemas, and inter-tool messaging protocols[cite: 1]. Break those, and systemic integration fails[cite: 1].
Everything else remains open to adaptation[cite: 1]. When an engineer discovers a better prompt pattern for their specific workflow, it is merged centrally so the entire organization benefits[cite: 1]. Our SiDekick skills layer grew out of this exact discipline[cite: 1]. In one recent instance, an engineer converted a rough Paint sketch into a polished network topology diagram for a client proposal in under an hour using a skill built and shared by a colleague[cite: 1].
Action Plan for Business Leaders
If your staff are deploying unmanaged AI tools across disparate departments, pause and establish a framework[cite: 1]:
- Define Core Context: Establish the baseline corporate context, data boundaries, and output formats every AI tool must inherit[cite: 1].
- Align Strategy First: Integrate AI governance directly into your broader managed IT strategy and planning roadmap[cite: 1].
- Host Alignment Sessions: Conduct focused internal workshops to align prompt expectations and eliminate fragmented workflows[cite: 1].
