Artificial intelligence is rapidly becoming the answer to almost every organisational question.
How do we improve productivity?
AI.
How do we make better use of our data?
AI.
How do we reduce costs, automate processes, improve customer experience or help people make better decisions?
Increasingly, the answer is AI.
And there are good reasons for the excitement. The capabilities developing around generative AI, automation and autonomous agents are extraordinary.
But there is a problem…
AI operates through the organisation that already exists.
If decision-making is unclear, AI doesn’t create accountability. If knowledge is fragmented, AI can surface conflicting versions of the truth faster.
If teams operate in silos, connecting them to increasingly sophisticated technology doesn’t suddenly make them collaborate.
And if a process shouldn’t exist in the first place, automating it simply creates a faster unnecessary process.
So perhaps before asking:
Where can we use AI?
We should ask a different question:
What kind of organisation are we about to amplify?
Efficiency isn’t the same as capability
Imagine an organisation where a relatively straightforward decision requires approval from seven people.
Someone proposes using AI to improve the process.
AI gathers the information required for the decision. It summarises the evidence, prepares the documentation and routes the request to each approver. It automatically chases people who haven’t responded and once everyone has approved it, the system creates a perfect audit trail.
The results look impressive, processing time falls and the administrative effort disappears. People spend less time chasing approvals.
The AI implementation is declared a success… but there’s a question nobody asked.
Why did seven people need to approve it in the first place?
The organisation has become considerably more efficient at executing a process that may never have needed to exist.
That’s the distinction between efficiency and capability:
Technology can help an organisation do what it already does faster. Transformation should change what the organisation is capable of doing.
Those aren’t necessarily the same thing.
AI can automate bureaucracy
This becomes particularly important as organisations move beyond individual AI tools towards automation and autonomous agents.
For years, many digital transformation programmes have digitised existing processes rather than questioning them.
The paper form became the digital form, the spreadsheet became the dashboard, the manual approval became the automated workflow.
The underlying organisational logic remained largely untouched. AI creates the possibility of taking this much further.
But that also creates a risk: a bureaucratic organisation with AI can become an extremely efficient bureaucracy.
A siloed organisation with AI can create more sophisticated silos.
An organisation with poor information can generate bad decisions considerably faster.
An organisation afraid to make decisions can acquire extraordinarily sophisticated information with which to continue avoiding them.
AI doesn’t automatically remove organisational dysfunction: it gives it leverage.
But perhaps that’s also the opportunity
There is another way of looking at this, however.
AI might become one of the most useful organisational diagnostic tools a business has ever had, because trying to automate work forces us to confront questions organisations have often avoided.
Who actually owns this process?
Why does this approval exist?
Where does this information come from?
Why are three teams maintaining different versions of the same data?
Who is allowed to make this decision?
Why does somebody manually copy information from one system into another?
What happens when two systems disagree?
Perhaps, most importantly:
Why do we do this at all?
These aren’t really technology questions - they’re questions about how the organisation works.
The technology simply makes them harder to ignore.
Technology reveals the organisation
This is why I increasingly think the most interesting thing about AI isn’t simply what it can automate - it’s what it reveals.
Every attempt to introduce AI into an organisation encounters the structures, behaviours, incentives and decisions that already exist.
Sometimes those structures are effective, sometimes they exist for good reasons and sometimes… nobody can quite remember why they’re there.
AI exposes all of it.
That means the friction encountered during AI adoption shouldn’t necessarily be treated as something standing in the way of transformation.
The friction may be telling us where the transformation actually needs to happen.
If an AI system can’t determine who owns a decision, perhaps the problem isn’t the AI.
If an agent can’t reliably execute a process because nobody agrees how the process works, perhaps we’ve learned something important.
If an AI assistant can’t answer a question because organisational knowledge is fragmented across dozens of systems and teams, perhaps the technology has revealed a problem that was already there.
So don’t fix the organisation first
None of this means organisations should wait until everything is perfectly designed before adopting AI.
They’d be waiting forever. Organisations are messy - they change and evolve, processes accumulate, responsibilities overlap, technology changes and people find workarounds - that’s normal.
The opportunity is to use AI implementation as a reason to question some of those assumptions.
Before automating a workflow, ask whether the workflow should exist.
Before introducing AI decision support, clarify who actually owns the decision.
Before connecting organisational knowledge, understand where authority and truth reside.
Before deploying autonomous agents, decide what autonomy really means.
Before celebrating improvements in productivity, ask what new organisational capability has actually been created.
Because the most successful organisations won’t simply bolt AI onto the way they already work.
They’ll use it as an opportunity to rethink how work gets done in the first place.
AI won’t fix a broken organisation.
But it might finally make the cracks impossible to ignore.
