AI promises speed. Efficiency. Automation. Yet, in many organisations adopting it today, the opposite is unfolding:
More approvals
More reviews
More reporting
More controls “just to be safe”
Instead of removing friction, AI often creates a new kind of friction, digital, distributed, and harder to see.
This is the fresh irony of transformation:
We automate the work… and then drown in the oversight.
Where Parkinson’s Law once expanded bureaucracy through headcount, AI risks expanding it through compliance workflows, audit trails, and policy gates.
If we’re not careful, the tools built to accelerate progress become the reason progress slows.
The New Source of Bureaucracy: AI Compliance
Regulations are fragmenting, rapidly. The ITU’s 2025 AI Governance Tracker notes:
65+ unique national AI regulations, all diverging
Mandated traceability even for low-risk automation
Escalating documentation for every model interaction
Organisations respond logically:
Create AI review boards
Appoint new “Responsible AI Leads”
Add human checkpoints where automation feels risky
All sensible. All adding drag.
Ethics becomes a bottleneck instead of a design principle.
A Case in Point
Across multiple governments experimenting with generative AI pilots, the outcomes have been similar:
Launch quickly for headline momentum
Hit legal and audit anxiety
Freeze or roll back due to accountability fears
Forethought’s 2025 analysis of North American public AI deployments found:
72% stalled before moving beyond proof of concept not because the tech failed, because governance wasn’t built for scale
It’s not the algorithm that breaks. It’s the approval process.
Bureaucracy Audit Matrix
Old drag vs. new drag, both equally dangerous
Parkinson warned that work expands to fill the time available.
In 2025, we add:
Controls expand to fill the uncertainty available.
As uncertainty grows, and AI introduces plenty, oversight expands even faster.
How to Break the Loop
AI should accelerate clarity, not oversight. A practical path:
1️⃣ Design accountability upfront One owner per workflow. Not a committee.
2️⃣ Classify risk by outcome, not technology Not all AI requires gold-standard governance.
3️⃣ Build automation with audits embedded Let traceability be an output of execution, not a separate task.
4️⃣ Measure friction Cycle time, review load, and decision latency must be tracked.
5️⃣ Keep humans for the art, automate the science If judgement isn’t required, don’t require it.
Governance should enable scale, not police it.
What This Makes Possible
When done right:
AI clears space for innovation
Compliance becomes proactive, not paralyzing
Human attention stays on value creation
And organisations stay alive to their purpose even as they grow.
The win is not more automation. The win is less work that slows us down.
Your Next Step, One Question
There’s a simple diagnostic to begin your audit:
Which AI success in your organisation quietly added a new layer of oversight?
Start there.
Fix that.
That’s how the real productivity gains begin.