Drift¶
Drift is an inherent property of dynamic systems. It is not an exception condition. It is what happens when systems run in changing environments with changing people, data, dependencies, policies, and incentives.
Drift becomes a governance concern when movement is not observed, understood, bounded, or corrected.
Foundational Premise¶
No dynamic system remains perfectly aligned indefinitely. Over time, systems naturally move structurally, operationally, strategically, culturally, informationally, and behaviourally.
Drift can emerge because:
- Environments change.
- Incentives evolve.
- Assumptions age.
- Unmanaged variability accumulates.
- Local optimisations distort global outcomes.
- People and organisations adapt imperfectly.
The role of leadership, architecture, governance, and maintenance is not to eliminate drift permanently. It is to detect drift early, maintain coherence, correct trajectory, preserve viability, and manage adaptation within acceptable bounds.
Organisational Drift¶
Organisations drift when decisions, incentives, priorities, handoffs, and accountability structures change faster than the operating model. A process may still appear intact while the meaning of roles, approvals, and success criteria has shifted.
Software Drift¶
Software drifts as dependencies change, assumptions age, interfaces evolve, configuration diverges, and usage patterns move beyond the original design. A system can keep passing basic checks while its actual behaviour becomes less aligned with its intended purpose.
Infrastructure Drift¶
Infrastructure drifts when deployed environments diverge from declared configuration, manual changes accumulate, runtime state differs from source control, and operational exceptions become permanent facts.
Prompt Drift¶
Prompt drift occurs when instructions, examples, system messages, retrieval context, or conversational history change the effective behaviour of an AI-driven workflow. The prompt may look stable while the active instruction environment is not.
Policy Drift¶
Policy drift occurs when written policy, interpreted policy, and executed policy diverge. This can happen when policies are ambiguous, outdated, unevenly enforced, or not represented in runtime systems.
Autonomous Drift¶
Autonomous drift occurs when an autonomous or semi-autonomous actor moves away from intended behaviour through local optimisation, partial context, tool effects, stale memory, delegation, changing incentives, or unobserved adaptation.
Agentic drift is an AI-specific subterm beneath autonomous drift. It is acceptable when discussing AI agents, but it should not be treated as the broader master term.
Why Drift Matters¶
Drift is not automatically bad. It can signal adaptation. The problem is unmanaged drift: movement without visibility, authority, constraint, or intervention.
Visible stability is not proof of internal coherence. A system may continue to produce acceptable surface signals while trust erodes, technical debt accumulates, incentives distort, alignment weakens, or resilience declines.
A drift-aware system asks:
- What changed?
- Who or what changed it?
- Was the change within authority?
- Did the change preserve intent?
- What trajectory does the change create?
- Which feedback confirmed the result?
- How long did correction take?
- What intervention is available if the change weakens coherence?