The Great Compression: Why AI Is Coming for the Layers Around Work
On organisational layers, what management actually does, and the capabilities that headcount reductions can hide.
In May 2026, Cloudflare announced it was cutting more than 1,100 jobs, roughly 20% of its workforce.
The announcement came alongside quarterly revenue of $639.8 million, up 34% from the previous year.
Its founders described the restructuring as part of a transition to an agentic AI-first operating model.
AI is beginning to eliminate some of the organisational machinery we've spent decades building around work itself.
The people who collect updates, reconcile reports, monitor activity, coordinate departments, translate information for leadership, and make sure everyone knows what everyone else is doing.
Not all of those functions will disappear. Many exist for good reasons. But the amount of human effort required to perform them is changing.
I've started thinking of this as the Great Compression: the shrinking distance between work being done, work being understood, and decisions being made.
We built organisations that need people to explain themselves to themselves
The bigger organisations become, the harder it is for anyone to understand what's actually happening inside them.
A CEO cannot personally know what thousands of employees are doing. Department heads cannot track every decision. Finance needs reconciliation. HR needs records. Operations needs reporting. Everyone needs meetings.
So companies build layers.
Analysts, coordinators, administrators, programme managers, reporting teams, auditors, and middle managers. People whose responsibilities include translating one part of the organisation into something another part can understand.
This was never inherently wasteful.
Most of these roles emerged because gathering, interpreting, and distributing information required significant human effort.
But AI changes the economics of that arrangement.
Software can increasingly monitor workflows, reconcile information, detect exceptions, generate reports, summarise conversations, and identify patterns without someone manually assembling the picture.
The information still needs to be accurate. Systems still make mistakes. Someone still needs to decide what deserves attention.
But the cost of producing visibility is falling.
How many organisational roles exist primarily because information used to be difficult to collect and distribute?
Management might have to become management again
Consider how much of a typical manager's week is spent gathering updates, passing information between teams, attending status meetings, escalating blockers, and explaining progress to someone further up the hierarchy.
In many organisations, we've quietly confused coordinating work with managing people.
If a system can tell leadership which projects are behind schedule, which teams are struggling, and where decisions are getting delayed, the manager whose primary contribution was producing that information becomes harder to justify.
But a good manager does considerably more than keep people informed.
They recognise potential before it becomes measurable. They understand why someone who performed brilliantly six months ago is suddenly struggling. They resolve conflict, develop judgment, create trust, and sometimes protect their teams from decisions that look perfectly reasonable in a spreadsheet.
AI can help with parts of that work. It cannot be assumed to understand all of it.
So I don't think AI necessarily makes management obsolete.
I think it makes management without much actual management increasingly difficult to defend.
And frankly, that distinction was overdue.
The dashboard doesn't know everything
As AI makes organisations easier to measure, leadership may begin to believe that organisations are also becoming easier to understand.
Those are different things.
Imagine an AI system identifies an employee whose productivity has dropped over three months.
The data might be perfectly accurate.
But what does it mean?
Perhaps they're disengaged. Perhaps they're handling the most difficult customers. Perhaps they've been helping struggling colleagues. Perhaps their tools are broken, their manager is ineffective, or the organisation is rewarding the wrong behaviour.
The system can identify a change in output. Understanding the reason requires context that may never appear in the data.
This is particularly interesting to me because I work in enterprise sales.
AI can already assist with account research, CRM updates, meeting preparation, prospecting, follow-ups, pipeline analysis, and a remarkable amount of the administrative work surrounding sales.
Much of that is useful. I'd happily have less administration in my working day.
But none of it automatically tells me whether a customer trusts me, whether a stakeholder is quietly resisting a decision, or whether the commercially sensible thing to do is to stop selling and listen.
And yet, those things can determine whether the sale happens at all.
A company may become better at seeing work while still being poor at seeing people.
That possibility concerns me more than the disappearance of reporting layers.
A redundant task doesn't necessarily mean a redundant person
Traditional workforce planning has largely revolved around job titles, departments, reporting structures, and headcount.
But jobs are collections of tasks, and people are more than their job descriptions.
Consider someone working in operations.
They might spend half their time producing reports, reconciling information, and coordinating approvals. AI could automate much of that.
But that same employee might also understand why a particular customer behaves differently, which internal processes routinely fail, which exceptions are genuinely dangerous, and how to get three departments to cooperate when the formal process breaks down.
That knowledge doesn't necessarily appear in the job description.
Remove the role without understanding the person, and the organisation may discover that it eliminated something considerably more valuable than the reports.
This is where I think workforce planning needs to change.
Instead of asking only how many employees a department needs, companies need to understand what work is actually being performed, which parts AI can take over, and what capabilities the people doing that work possess.
Some of those capabilities will remain valuable in entirely different roles.
Others may become more important precisely because routine work has been automated.
And some will genuinely become less valuable.
There is no reason to pretend every role can be preserved. But there is equally little reason to assume an organisational chart tells us everything worth knowing about the people on it.
A smaller organisation isn't necessarily a better one
There are two broad ways companies can respond to this compression.
The first is straightforward: use AI to reduce headcount, consolidate teams, remove management layers, and lower operating costs.
The other is to use AI to eliminate work that doesn't need human attention, then rethink where that attention should go.
Move people away from repetitive coordination and towards problems requiring judgment, relationships, creativity, expertise, and accountability.
A company can reduce its workforce by 20% and become more efficient. It can also reduce its workforce by 20% and lose institutional knowledge, customer relationships, leadership capacity, and abilities it won't realise it needed until something goes wrong.
The headcount reduction is easy to measure.
The capability loss may take years to become visible.
The real test won't be how much of the organisation they can compress. It will be whether they understand what they're compressing.
Because when a role disappears, the tasks don't always disappear with it. And neither does the need for the judgment, experience, and relationships that might have lived inside it.
Those are considerably harder to automate.
And much easier to lose.
Originally published on LinkedIn on May 22, 2026.
Sources: Cloudflare's announcement and Q1 2026 financial results.