AI is sold as a way to go faster and do more with less. In a lot of organisations adopting it right now, the opposite is happening. More approvals. More reviews. More reporting. More controls added just to be safe.
Instead of removing friction, AI often creates a new kind. It is spread across the org and harder to see than the old kind.
So you get a strange result. We automate the work and then spend the time we saved on oversight.
Parkinson's Law expanded bureaucracy through headcount. AI expands it through compliance workflows, audit trails and policy gates. If we are not careful, the tools we built to go faster become the reason things slow down.
The new source of bureaucracy: AI compliance
Regulation is fragmenting fast. The ITU's 2025 AI Governance Tracker records 65+ unique national AI regulations, all diverging from each other. It points to mandated traceability even for low-risk automation, and to escalating documentation for every model interaction.
Organisations respond in a way that makes sense on paper. They create AI review boards. They appoint a Responsible AI Lead. They add a human checkpoint wherever automation feels risky.
Each step is sensible. Each step adds drag. Ethics review becomes a bottleneck rather than something built in early.
A case in point
Across governments running generative AI pilots, the pattern is the same. They launch fast to get the headline, then hit legal and audit anxiety, then freeze or roll back because nobody wants to own the accountability.
Forethought's 2025 analysis of North American public AI deployments found 72 per cent stalled before moving past proof of concept. Not because the technology failed, but because the governance was never built to scale.
The technology works. The approval process is what stalls.
Old drag and new drag
Parkinson warned that work expands to fill the time available. The same thing now happens with controls: they expand to fill the uncertainty available. As uncertainty grows, and AI creates plenty of it, oversight grows faster still.
How to break the loop
AI should buy you clarity, not more oversight. A practical path:
1. Design accountability upfront
One owner per workflow. Not a committee.
2. Classify risk by outcome, not technology
Not all AI needs gold-standard governance.
3. Build automation with audits embedded
Let traceability fall out of execution instead of being a separate task.
4. Measure friction
Track cycle time, review load and decision latency.
5. Keep people on the judgement calls; automate the routine
If judgement is not required, do not require it.
Governance should enable scale, not police it.
What this makes possible
Done right, AI clears space for new work. Compliance becomes proactive rather than paralysing. Human attention stays on the things that create value. And the organisation stays close to its purpose even as it grows.
The gain is not more automation. It is less work that gets in the way.
Your next step: one question
Start the audit with one question. Ask yourself which AI success in your organisation quietly added a new layer of oversight.
Start there. Fix that. That is where the real productivity gains begin.
AI's real drag is not the technology but the oversight it triggers. Assign one owner per workflow, classify risk by outcome, and let audits fall out of execution so governance enables scale instead of policing it.