The Divergence
2026 isn’t the year AI slows down. It’s the year it splits.
Last year was about unhobbling — removing the practical constraints that stopped AI from delivering value. Progress didn’t come from breakthroughs. It came from connecting what already existed: better context handling, cleaner integrations, agents that could actually run inside real systems. Quiet progress, but real progress.
That phase is ongoing, but at the same time something different is now taking shape.
The field is moving in several directions at once. Some teams are pushing models into new conceptual territory; others are refining what we already have. Some organisations are rebuilding workflows for autonomy; others are still bolting chat interfaces onto legacy systems. Some people see AI as an amplifier; others see risk and decline. The gap between these approaches is widening, and the choices made in the next twelve months will matter far more than they did last year.
This report is about those forks — the places where technology, organisations, culture, and behaviour pull in different directions — and what they tell us about the year ahead.
Bits and Gambits
Five forks in the road. Some technical, some political, some bordering on spiritual. Together, they'll determine not just how AI develops in 2026, but who controls it and what it does to us in return.
The Geopolitical Fork: Diffusion vs. Superintelligence
The world is splitting into two distinct AI races. While America obsesses over superintelligence — racing toward artificial general intelligence and pouring resources into frontier models — China has pivoted to something fundamentally different: diffusion over dominance.
According to China's forthcoming five-year plan, the priority isn't beating the West to AGI but integrating AI into every sector of the economy through a campaign called "AI+". By 2027, they aim for widespread adoption across industry, healthcare, education, and government. By 2030, AI should be as ubiquitous as electricity.
The Technological Fork: Refining vs. Reinventing
The plateauing of large language models has split the industry between those who think transformers still have room to run and those betting on something fundamentally new. Yann LeCun champions "World Models" that learn by understanding causality rather than pattern-matching text.
DeepSeek's recent research on training stability suggests the transformer architecture isn't exhausted yet — their "Manifold-Constrained Hyper-Connections" method enables richer internal communication while preserving efficiency. At the same time, recursive language models are making waves, offering an alternative to the brute force approach to context models.
The Enterprise Fork: Bolting On vs. Rewiring
The corporate world is splitting between companies bolting chatbots onto legacy systems and those rewiring themselves around agentic workflows. The former is comfortable and incremental — a new interface for old processes. The latter is disruptive, requiring redesigned workflows and acceptance that automation will eliminate roles while creating new ones.
The data increasingly validates the latter approach: 88% of early AI agent adopters are now seeing positive ROI on their generative AI investments. Companies doing the rewiring will look unrecognisable within a few years, those that don’t likely won’t be around.
The Societal Fork: Loving Grace vs. Everyone Dies
Two visions of AI's impact are crystallising. One sees AI as amplifying human capabilities and solving problems at scale. The other takes catastrophic outcomes seriously. Books like If Anyone Builds It, Everyone Dies have gained traction for a reason.
The fork isn't just about which vision proves correct, but which shapes policy and investment. Optimism means lighter regulation and faster deployment. Pessimism means moratoriums and stricter controls. Neither is obviously preferable.
The Individual Fork: Watching vs. Doing
As AI makes everything available on demand, culture atomises. Personalised content means fewer shared reference points, less common ground. Simultaneously, the AI adoption gap widens between those who learn to actively use these tools and those who remain passive consumers. It's the difference between being the human in the loop and being the human left out.
These forks aren't independent. They reinforce and complicate each other. Geopolitical competition drives technological divergence. Technological choices shape enterprise adoption. Societal anxiety influences regulatory environments. Individual agency determines who benefits.
Those are the overarching trends, what follows are twelve specific trends that will also feature in the year ahead.
A Defining
Year Ahead
2026 is shaping up to be a hinge moment. After a year of connectors, protocols and quiet rewiring, the centre of gravity is shifting again. New models are arriving. New methods are emerging. Whole categories of work — from robotics to agentic systems to simulation — are moving from research to deployment.
The aim of this report has been to show the shape of that shift: the forks forming in technology, enterprise, society and culture — and the opportunities that come with them. We aren’t heading into a quieter year. We’re heading into one where the foundations for the next decade are set.