The Machine Needs a North Star

Meta’s faltering attempt to become an “AI-native” company reveals something much bigger than the limitations of today’s artificial intelligence. It raises a fundamental question about what an organisation is for, and why purpose will become more important as machines take on more of the work.

Within the space of a few days last week, two stories emerged about Meta — Mark Zuckerberg’s technology giant and the owner of Facebook, Instagram and WhatsApp — that appeared, at first glance, to have very little to do with one another.

The first dominated the headlines. Meta agreed to pay up to $18 billion over the next decade to settle claims brought by US states alleging that Facebook and Instagram had been designed in ways that harmed and addicted children. Meta admitted no wrongdoing, but the agreement reflects a much wider international reckoning over social media and its impact on young people.

The second story received less attention, but may ultimately prove just as consequential. In a special report Reuters revealed that inside Meta, Zuckerberg and his senior team had been attempting something extraordinarily ambitious. Not simply introducing artificial intelligence into the organisation, but redesigning the entire organisation around it.

Its codename was Project OT (Organization Transformation). The ambition was to make Meta “AI native”. Traditional product teams of ten or twenty people could become pods of three to five “builders”. Specialist roles would blur, management layers would disappear and AI agents would take on increasing amounts of work. Some planning scenarios contemplated reducing the size of individual teams by as much as 60%.

But then things went quickly wrong. Employees rebelled, morale plummeted and plans for a second wave of restructuring were abandoned. More importantly, Meta’s own internal evidence suggested that the promised productivity revolution was proving far more complicated than anticipated. According to Reuters, code changes to Meta’s internal systems increased by 220% year on year, while changes resulting in new or improved features actually reaching users rose by only 36%. Technical and security incidents reportedly increased by 40%.

There is an important lesson buried inside those numbers: More output is not the same thing as more value. And that fact takes us far beyond Meta.

For more than a century, management thinkers have wrestled with the question of how organisations should be structured and what they should exist to achieve. The twentieth-century corporation was largely hierarchical. Authority travelled downwards, information upwards and managers coordinated people occupying clearly defined roles. In business, Milton Friedman’s argument that the responsibility of business was to increase profits provided an equally clear organising logic.

Other models followed. Peter Drucker shifted attention towards customers and knowledge workers. Charles Handy repeatedly challenged conventional assumptions about organisations, hierarchy and the nature of work, anticipating a world in which organisations would become less rigid, careers less linear and knowledge increasingly more important than physical assets. Stakeholder theory widened the responsibilities of business, reflecting a growing recognition that organisations exist within complex networks of relationships rather than in isolation. Agile methodologies challenged rigid planning, while Frédéric Laloux’s book Reinventing Organizations proposed self-management and “evolutionary purpose”.

My own small contribution to this continuing conversation came with my book CORE in 2017, when I introduced the Single Organizing Idea (SOI®) to a wider audience and the idea that organisations perform better when governance, strategy, culture, innovation, operations and performance are aligned around a clear and shared purpose that is both valuable and relevant to all stakeholders.

The important word has always been ‘organising’, however you chose to spell it. Purpose was never intended to be something inspirational sitting above the real business of the organisation. Properly understood, it should help organise it. This is why Meta’s curtailed ‘Organizational Transformation’ experiment matters.

AI is not simply another office technology. The spreadsheet made accountants faster. Email transformed communication. The internet radically reduced the cost of distributing information. AI agents may change something much deeper. The way organisations themselves are organised.

If organisations can deploy hundreds or thousands of digital agents capable of researching, analysing, coding, communicating and undertaking increasingly complex sequences of work, many of the structures we have traditionally used to organise human activity begin to look very different. Teams may become smaller, job descriptions broader, management layers thinner and organisational boundaries more fluid.

Meta may have moved too quickly, but that does not necessarily mean it is wrong about the direction of travel. The harder question is what happens if this becomes normal.

Imagine an organisation of 500 people working alongside 5,000 AI agents, all constantly analysing information, initiating tasks and making decisions. No chief executive could instruct them individually and no operating manual could anticipate every circumstance. If the organisation changes, what organises the organisation?

This is where much of the current discussion about AI begins at the wrong end. We ask what AI can do, which jobs will disappear, how many employees companies will need and how much productivity can be extracted. Those are questions about means. But the more important questions concern ends.

Faster towards what? More efficient at doing what? More productive in whose interests? Better according to whose definition of better?

Artificial intelligence is powerful because it can optimise. Give it an objective and it can increasingly find efficient ways of pursuing it. But optimisation requires orientation. And at this point the other Meta story becomes relevant.

For years Facebook and Instagram became extraordinarily sophisticated at maximising engagement. Their systems learned how to keep people looking, clicking and returning. Whatever one’s position on the litigation, the scale of the settlement illustrates that extraordinary optimisation can produce consequences far beyond the metric being optimised.

Society is now asking whether maximising ever more human attention was always the right objective in the first place. That question cannot be answered by a better algorithm. It is a question of governance and ultimately purpose.

This is why the two Meta stories belong together. One concerns the consequences of technology Meta has already built. The other concerns how Meta intends to organise itself to build the next generation. Both lead to the same question: What is all this extraordinary capability actually for?

AI may reduce the need for hierarchy, but it will not remove the need for coherence. In fact, it will likely increase it. Organisations composed of humans working alongside intelligent agents will require a clear shared understanding of the outcome they exist to create and the principles governing how that outcome should be pursued.

That is precisely what genuine organisational purpose should provide. A Single Organizing Idea becomes an organisational compass, something against which boards can govern, leaders can develop strategy, teams can make decisions and, increasingly, AI systems can be designed and directed.

Not because machines should determine an organisation’s purpose. Quite the reverse. Humans must determine the purpose precisely because machines will increasingly determine how that purpose is pursued.

This is why the emerging ISO 37011 guidance on the governance of organisations by purpose feels so timely and necessary. The central governance challenge of the AI age may not simply be deciding where artificial intelligence can be deployed, but ensuring that extraordinary new capability remains directed towards outcomes worth pursuing.

There is a danger that organisations will see AI primarily as an opportunity to become faster, leaner and cheaper. But a badly oriented organisation does not become a better organisation simply because it becomes more efficient. It simply becomes a badly oriented organisation capable of travelling considerably faster, and the likely consequences of that should worry us all.

Reuters’ account offers an early warning. The future organisation will not simply require better artificial intelligence. It will require better human intention. Because the more powerful the machine becomes, the more important the North Star and the power of purpose.