AI in 2026: Current Trends and What Could Come Next

When we last looked at current trends in artificial intelligence in 2024, generative AI was experiencing a period of rapid development. Large Language Models (LLMs), Natural Language Processing (NLP), increasingly autonomous systems and more personalised AI were beginning to change how people interacted with technology.

We also explored the distinction between Artificial Narrow Intelligence (ANI), Artificial General Intelligence (AGI) and Artificial Superintelligence (ASI), and considered how advances in AI could eventually move us beyond the narrow systems we use today.

Only two years later, parts of that discussion already feel surprisingly distant.

Generative AI is no longer primarily associated with asking a chatbot a question or generating an image from a prompt. AI systems are increasingly able to reason across more complex problems, work with text, images, audio and video, use software, generate code, analyse large volumes of information and carry out sequences of tasks with less direct human input.

At the same time, many of the limitations we were dealing with in 2024 have not disappeared. AI can still hallucinate, misunderstand context and produce incorrect information with considerable confidence. Questions around security, copyright, data governance, accountability and human oversight have also become more prominent as organisations find more ways to deploy the technology.

So, where does AI actually stand in 2026 – and what might the next few years look like?

AI in 2026: More capable, more widely used – but still imperfect

Perhaps the biggest change since 2024 is not simply that AI has become more powerful. It has become considerably more commonplace.

Generative AI tools have moved beyond their early novelty and are now being used for everything from research and writing to software development, data analysis and personal organisation. Harvard Business Review's 2026 study of how people are using AI found an increasingly broad range of applications, alongside developments such as the growth of "vibe coding" and agentic workflows.

The scale of organisational adoption has grown as well. Stanford University's 2026 AI Index Report found that organisational AI adoption had reached 88%, while generative AI reached an estimated 53% population adoption within three years – considerably faster than the historical adoption of technologies such as the PC and the internet.

The capabilities behind these systems have advanced just as quickly.

From chatbots to AI agents

One of the most important AI trends in 2026 is the development of AI agents.

Traditional generative AI largely responds to individual prompts. Agentic AI goes further by allowing systems to work towards a broader objective, break that objective into steps, use software or other digital resources and carry out multiple actions before returning a result.

This does not mean autonomous AI systems can simply be trusted to manage complex processes without supervision.

Stanford's AI Index illustrates what researchers sometimes describe as AI's "jagged frontier". Modern systems can perform extraordinarily well at certain highly complex tasks while continuing to struggle with tasks humans might consider comparatively straightforward.

In 2026, leading AI systems were capable of gold-medal-level performance in International Mathematical Olympiad problems. Yet the strongest model tested by Stanford could correctly interpret an analogue clock only around half of the time. Meanwhile, AI agents improved from approximately 12% to 66% task success on OSWorld, a benchmark based on performing real computer tasks – an enormous improvement, but still a failure rate of roughly one in three attempts.

That contradiction is important. The future of AI is unlikely to be defined simply by whether systems become "intelligent enough". Reliability, context, permissions and the consequences of getting something wrong will determine how much autonomy organisations are actually prepared to give them.

Multimodal and more specialised AI

AI is also becoming less dependent on a single type of information.

Multimodal AI can work across combinations of text, images, audio, video and other forms of data. Rather than treating each format as an isolated input, increasingly capable systems can interpret information across several formats within the same interaction.

This creates opportunities well beyond conversational AI. Systems could, for example, analyse an image while considering accompanying written information, interpret spoken instructions and visual surroundings together, or combine operational data from several sources when assisting with a decision.

IBM notes growing interest in smaller, more efficient models alongside systems trained or adapted for particular industries and applications. These models can require fewer computing resources and can be better suited to specialised enterprise environments where organisations need AI to work within a particular body of information or clearly defined processes.

Over time, this could mean the AI people interact with most often is not necessarily a single general-purpose chatbot. It may increasingly be AI embedded within the software, devices and systems they already use.

AI in 2027: How quickly could things change?

If forecasting AI was difficult in 2024, it has arguably become even harder.

One of the more widely discussed attempts to imagine the immediate future is AI 2027, a scenario published in 2025 by Daniel Kokotajlo, Scott Alexander, Thomas Larsen, Eli Lifland and Romeo Dean.

The scenario explores an extremely rapid trajectory in which increasingly capable AI agents begin contributing substantially to AI research itself, accelerating subsequent development and ultimately leading towards superhuman systems.

Its authors describe the scenario as their "best guess" of what such a future could look like, informed by trend extrapolation, expert feedback and forecasting exercises. At the same time, they explicitly encourage alternative scenarios and acknowledge that uncertainty increases considerably as their forecast progresses, as such, AI 2027 acts more as “potential trajectory”.

AI forecasting currently covers an unusually wide range of outcomes. Some researchers expect progress to continue at or above its present pace. Others expect technical, economic, energy, data or deployment constraints to slow development. There is also continuing disagreement around what should constitute AGI in the first place.

Stanford's 2026 report found that AI capability is not currently plateauing, with substantial improvements appearing across coding, reasoning, science and agentic benchmarks.

But rapid progress in benchmarks does not automatically translate into equally rapid transformation everywhere else.

Integrating AI into organisations requires reliable data, appropriate infrastructure, security controls, changes to processes and trust. A system that succeeds 90% of the time may be highly useful in one environment and completely unsuitable in another where the remaining 10% could have serious consequences.

The development of AI and the adoption of AI are related, but they do not necessarily move at the same speed.

What could AI look like by 2030?

A more useful way to think about the next several years may therefore be to look at a range of possible outcomes rather than attaching artificial certainty to a single prediction.

In June 2026, the UK Government Office for Science published an updated set of AI Scenarios 2030, developed with input from government, academia and industry.

The report deliberately considers several different trajectories. At one end is Slow Burn, in which capability advances more slowly and AI causes less disruption than anticipated. At the other is Take-Off, where AI exceeds expert human performance across virtually all cognitive tasks and development accelerates dramatically. Between these sit scenarios including Augmented Growth, where highly capable AI remains largely complementary to human work, and Transformation Economy, where extensive automation produces considerably greater labour disruption.

The Government Office for Science makes an important distinction: these scenarios are not predictions. The real future may contain elements of several of them.

That may be the most realistic way to approach AI in 2030.

We cannot confidently say whether AGI will exist by then, whether progress will slow or whether a major technical breakthrough will accelerate it further. We can, however, identify several developments that are already underway.

AI is likely to become more embedded. Instead of AI always being something users consciously open and prompt, more systems are likely to incorporate AI within everyday software and operational processes. AI may increasingly sit behind search, analysis, scheduling, decision support, customer interactions and other functions without necessarily being presented as a standalone product.

AI agents are likely to become more capable. The agentic systems available in 2026 remain inconsistent, but the improvement in their ability to interact with computers and complete multi-stage tasks is substantial. By 2030, organisations may delegate increasingly complex processes to AI while defining limits around what those systems can access, decide and execute.

Specialised AI is likely to grow alongside general-purpose models. Industry-specific and organisation-specific systems can provide advantages where context, security and domain knowledge matter. Smaller models may also allow more AI processing to take place locally or directly on devices rather than relying entirely on large cloud-based systems.

The relationship between people and AI will continue to change. Automation will undoubtedly affect some tasks and occupations, but the extent of displacement remains uncertain. One of the UK Government's 2030 scenarios, Augmented Growth, considers a future in which AI reaches very high levels of capability while humans remain involved in most activities and new roles emerge alongside the technology. Other scenarios anticipate considerably greater disruption.

Rather than a simple choice between humans or AI, the more immediate question for many organisations is likely to be which decisions should remain human, which tasks can be automated and where the combination of human judgement and AI capability produces better results.

And, as AI takes on more responsibility, governance is likely to become increasingly important. As AI systems become more autonomous, organisations will have to consider not only what AI can do, but what it should be allowed to do, what information it can access, how decisions are audited and where accountability ultimately rests.

When we looked at AI trends in 2024, much of the discussion centred on the rapid development of generative AI and whether increasingly capable narrow AI systems could eventually lead towards Artificial General Intelligence.

Two years later, that question has not disappeared.

If anything, advances in reasoning, multimodal AI and autonomous agents have made the possibility of increasingly general systems a more prominent part of the conversation.

But developments since 2024 have also demonstrated why the future of AI cannot easily be reduced to a single timeline.

Today's systems can complete tasks that would have seemed remarkable only a few years ago - while continuing to make surprisingly basic mistakes. Adoption is accelerating, yet reliability, security, governance and human oversight remain important constraints. AI can complement people in one setting, automate their work in another and remain unsuitable for deployment altogether in a third.

By 2030, AI will almost certainly be more capable, more autonomous and more deeply embedded within the technology and systems we use.

Whether that represents a continued evolution of today's AI, the emergence of something that could reasonably be called AGI, or another development we have not yet anticipated is much harder to predict.

For now, perhaps the safest prediction about the next five years of AI is that change will continue – but it is worth remaining realistic about both its opportunities and its limitations.