AI is starting to redraw the org chart: four companies changing how work gets done

In the space of three days, four very different companies gave a surprisingly coherent picture of where workplace AI may be heading next.

On 29 September, Millennium described how it is giving employees personalised AI “Digital Twins”. A day later, BMW linked much wider use of AI with leaner management structures. The same day, Whitbread explained how one of its transformation teams is already treating AI as a “third workforce”. Then, on 1 October, Japanese software company HENNGE launched a subsidiary designed to operate with two directors and no dedicated employees.

These are not four versions of the same model. But together they point to a shift that matters for HR: companies are starting to make decisions about structure, capacity and accountability on the assumption that some work will be carried out by AI agents rather than people.

HENNGE: what if you design the company around AI from day one?

The clearest experiment comes from Japanese SaaS business HENNGE.

Its newly created subsidiary, HENNGE AI, has two directors and no dedicated employees. The company says AI agents will perform software development, customer support, marketing and back-office work. Sales and selected corporate functions will still be supplied by the parent company.

So this is not a completely autonomous company: human directors remain accountable and the business still depends on the wider HENNGE group. Even so, HENNGE is not taking an existing team and asking where AI can save time. It is asking what an organisation should look like if AI agents are built into the operating model from the beginning.

Representative director Takeru Imaizumi describes HENNGE AI as its own “proof of concept”. The experiment should expose practical questions about supervision, security, cost and ownership that are much harder to see in an AI strategy deck.

BMW: if management layers shrink, where does the management work go?

BMW is redesigning an established organisation rather than creating a new one.

At its September Capital Market Day, the carmaker said it intends to reduce the number of divisions and associated management roles by 20% by the middle of 2027, with comparable reductions at the organisational levels below.

In the same plan, BMW positioned wider use of AI as a route to faster decisions, greater efficiency and leaner structures, with agents supporting work across the business.

It would be too simple to say AI is directly replacing 20% of BMW’s managers. The restructuring sits inside a much broader cost and simplification programme.

But the connection BMW itself is making is important. If an organisation removes management layers while asking AI to absorb more coordination, analysis and routine activity, the question is no longer just how individual managers use AI. It is which parts of management work still require a manager at all, and who takes on the judgement, coaching and escalation work when a layer disappears.

Millennium: when every employee can have a digital counterpart

Millennium is changing the unit of work in a different way.

The investment firm has made a personalised Digital Twin available to every employee who wants one. More than 1,600 were already in use when the firm published details of the programme in September.

Each twin works for one employee. It can retain preferences and context, handle recurring research, prepare meeting materials, monitor information and run scheduled tasks overnight. It has its own identity, email address, permissions and audit trail, while the employee controls its access to systems.

Millennium is also developing AI “Teammates” for teams and business functions.

That begins to blur a familiar boundary in job design. What happens when one employee routinely delegates part of their role to a persistent digital counterpart that remembers context and has access to company systems? The employee remains accountable, but may also need to configure, direct and check another source of capacity attached to the role.

Whitbread: should workforce planning extend beyond headcount?

Whitbread offers perhaps the most immediately transferable example for HR.

The Premier Inn owner is expanding Microsoft Copilot from a 270-person pilot to 1,650 employees and has created an AI Enablement Office to govern how employee-built agents move from individual experiments into shared company tools.

During the pilot, employees developed more than 60 role-specific use cases across eight functions. More interestingly, Emma Cluett, Head of Delivery Transformation, said her team had remained roughly the same size because Copilot was being used as a “third workforce”.

That framing pushes AI into workforce planning. Organisations are used to thinking about permanent employees, contractors, consultants and outsourced providers as different ways of supplying capacity. Whitbread is beginning to describe AI in similar terms.

The productivity evidence should be treated with some caution: the detailed case study is published by Microsoft and the time savings are company-reported. But if AI can provide useful capacity without adding headcount, workforce planning may increasingly need to account for what is performed by people, what is delegated to agents and where additional human capacity is still required.

The common shift: from AI adoption to organisation design

HENNGE is testing an almost employee-free operating company. BMW is flattening an existing hierarchy. Millennium is attaching AI capacity to individuals and teams. Whitbread is beginning to treat AI as another source of delivery capacity.

The common thread is that AI is moving beyond helping somebody do their existing job faster.

That same shift appeared in Vertage’s recent move from headcount to outcomes, where the workforce group described a model in which organisations define the work first and then decide whether the capability should be built, bought, borrowed or delivered by a bot.

For HR, previously theoretical questions are becoming operational. Who is accountable for work carried out by an agent? How should access and identity be managed for non-human workers? How should capacity be measured when not all of it comes from people? And when AI changes the amount of work one role can absorb, when should the job itself be redesigned?

But that is precisely why these experiments are worth watching. The AI-native organisation may not arrive as a single dramatic reinvention. It may emerge through a series of smaller structural choices: one management layer removed, one agent given an identity, one team held at the same size, or one new company built without employees.