This article is a forecasting exercise: we build plausible scenarios from current research. No crystal ball. Every assumption is sourced and qualified. Enjoy the read.
The current picture: where does AI really stand in 2026?
To anticipate what comes next, we first need to take stock of where things stand today. Language models (LLMs: Large Language Models, these AIs trained on enormous quantities of text to generate and understand human language) have reached a level that surpasses the average human on most standard intellectual tasks. Claude Opus 4.7 (the latest generation of Anthropic's model), GPT-5.5 (OpenAI's most advanced version) and Gemini 3.1 Pro (Google's frontier model) handle complex reasoning, code, creation and analysis at a level that would have seemed impossible three years ago. As we documented in our article on unmanaged acceleration, even the engineers themselves cannot always predict what a model will do before testing it.
Adoption is exploding: 16.3% of the world's population uses generative AI regularly according to Microsoft, with peaks of 64% in the Emirates. Autonomous agents (AI programs capable of executing sequences of tasks without human intervention) are starting to carry out complex tasks in the enterprise, as we described in our article on the startups that outperform. Humanoid robotics is moving out of the prototype stage and into production (Boston Dynamics Atlas, Figure 02). And AI is accelerating scientific discovery in several fields, notably biology (AlphaFold 3, DeepMind's model that predicts the 3D structure of proteins, has mapped 200 million of them).
It is from this base that the following scenarios are built.
One-year horizon (May 2027): what is almost certain
At this horizon, we are in the realm of near-mechanical prediction. AI labs have public roadmaps, announced capabilities will ship (with a few delays), and uses will spread massively.
First expected change: AI agents will be fully operational in the enterprise. According to Gartner (one of the leading global technology market research firms), 40% of enterprise applications will contain task-specific agents by the end of 2026, up from less than 5% a year earlier. In practice, this means that repetitive administrative tasks (invoice entry, email triage, scheduling appointments, report generation) will be massively automated in 2027 across most large companies.
Second change: humanoid robotics will cross the threshold into limited mainstream adoption. Boston Dynamics (the American robotics pioneer, acquired by Hyundai), Figure (a start-up making humanoid robots for industry) and Tesla (with its Optimus robot) have all announced commercial deployments for 2026-2027. We are talking about robots costing $20,000 to $50,000, sold first to industrial companies, then to wealthy individuals in 2027. Capabilities will remain limited (simple manipulation, navigation in controlled environments), but the psychological barrier will have been crossed.
Third change: generative video and 3D models will reach "professional production" standard. Veo 4 (Google's video model), Seedance 3 (ByteDance's video model), or a successor to Sora (probably more profitable this time) will make it possible to produce adverts, short films and complete marketing content in a matter of hours, where entire teams were once needed. As we analysed in our article on the disruption facing graphic designers, this transformation is already under way and will accelerate.
The most worrying scenario at this horizon is not technological but social. According to figures from Meta, Microsoft and Amazon, more than 150,000 tech jobs have already been cut since January 2026, largely justified by AI automation. If this pace continues, we are talking about 300,000 to 500,000 cumulative cuts by mid-2027. European companies, with stronger social protections, will slow this trend, but they will not be able to avoid it indefinitely.
Five-year horizon (2031): the likely leap
At five years, things get interesting. Several major bets are in play, and today's scientific advances are beginning to produce commercial effects.
Hypothesis 1: AGI will emerge between 2027 and 2031. AGI (Artificial General Intelligence) refers to an AI capable of matching or surpassing humans on most intellectual tasks, not just on specific ones. It is the most controversial prediction of the moment. Dario Amodei (CEO and co-founder of Anthropic, former vice-president of research at OpenAI) speaks of "a country of geniuses in a data centre" as early as 2027. Sam Altman (CEO of OpenAI) points instead to 2030-2035. Surveys of 8,500 academic researchers give a median of 2040-2061. Ben Goertzel (CEO of SingularityNET, one of the historic pioneers of the field and the man who popularised the term AGI in the 2000s), is aiming for 2027-2029.
The gap between the predictions of commercial lab CEOs (2026-2031) and those of academic researchers (2040-2061) is enormous. Part of this gap is probably commercial bias (CEOs have an interest in promising fast to raise capital). But part of it also reflects the fact that the labs see what is happening internally and that capabilities are progressing faster than scientific publications allow us to see. Reasonable probability: 30 to 50% for a functional AGI by 2031. Not a reassuring figure, nor an alarming one. Just realistic.
Hypothesis 2: widespread personalised medicine. Insilico Medicine (a Hong Kong-based company that uses AI to discover drugs, valued at several billion and publicly listed) is one of dozens of companies using AI to discover drugs. AlphaFold 3 predicts the structure of all known proteins. The compression of the drug development cycle (from 10-15 years today to 3-5 years in 2031) is likely. For patients, this means the arrival of treatments for orphan diseases, specific cancers, and probably the first cellular anti-ageing treatments.
Hypothesis 3: the rationalisation of the AI market. As we analysed in our article on the AI bubble, the market will probably go through a painful consolidation. By 2031, we should expect 3 to 5 dominant players per segment, with a concentration of value similar to what happened in the cloud (Amazon's AWS, Microsoft's Azure, Google's GCP). Users will probably pay more for more mature and more stable services.
10-year horizon (2036): structural transformations
At ten years, we enter territory where caution is called for, but where certain trajectories are already under way and difficult to reverse.
Medical scenario: healthy life expectancy could reach 90-100 years for those with access to care. The anti-ageing market should exceed $420 billion by 2030, with annual growth of 8-9%. Insilico Medicine, Altos Labs (a California lab founded in 2021 and funded by Jeff Bezos that works on cellular rejuvenation), Calico (a longevity lab created by Google in 2013), and dozens of others are actively working on cellular ageing. The "Horvath Clocks" (epigenetic clocks developed by researcher Steve Horvath, which measure a person's biological age from DNA markers, distinct from chronological age) already make it possible to measure the effectiveness of individualised protocols. By 2036, several anti-ageing therapies will probably have passed phase 2 and phase 3 clinical trials (stages of human clinical trials, more advanced and necessary to obtain marketing authorisation).
Note: life expectancy does not become infinite, and the promises of transhumanists (a movement that advocates enhancing human capabilities through technology, sometimes aiming for immortality) are not validated by academic medicine. But living in good health to 100 for a growing share of the population in rich countries is now a credible scenario.
| Area | Today (2026) | 2036 Scenario |
|---|---|---|
| Healthy life expectancy (rich countries) | 70 years | 85-95 years |
| Drug development cycle | 10-15 years | 2-3 years |
| Cancers treatable with 5-year+ survival | ~70% | ~90% |
| Clean energy in the global mix | ~30% | 50-60% |
| Jobs transformed by AI (% of employment) | 15% | 60-80% |
Economic scenario: the profound transformation of work. By 2036, the question will no longer be "which jobs will disappear" but "how societies managed the transition". The countries that put in place a universal basic income early (payment of a regular sum to all citizens without a means test) or a transition income (Finland, the Emirates, Singapore) will probably be the most stable. Those that let the market adjust brutally (the United States under certain administrations) will be the ones where inequality exploded.
To borrow Sam Altman's words: "Intelligence is going to become broadly accessible, as electricity was in the 20th century". If that is true, economic value will shift towards what cannot be duplicated: unique creativity, human bonds, the physical experience of the world. Care work, craft trades and human support roles could gain in relative social value.
Space scenario: the first lunar and Martian bases will begin. NASA has an "AI Track 2040" strategy that banks on autonomous exploration accelerated by AI. The Perseverance rover (the exploration vehicle sent to Mars by NASA in 2020) already uses AI-planned trajectories. By 2036, we can reasonably anticipate: a human return to the Moon (probably before 2030 with Artemis, the American programme to return to the Moon), the first permanent lunar bases (2032-2035), the first human mission to Mars (around 2035-2040). "Crewed" space exploration finally becomes concrete, not thanks to Elon Musk alone, but thanks to the acceleration AI brings to trajectory calculations, the design of radiation-resistant materials, and precursor robotics.
The 30-year horizon (2056): the major forks
At thirty years, we are in realistic science fiction. Scenarios diverge, and the future will depend on political and economic choices made right now.
Optimistic scenario: the successful transition. Human civilisations have managed to integrate AI without a major crisis. Healthy life expectancy reaches 110-120 years in developed countries. Mars is colonised by several thousand people. Nuclear fusion (a process that mimics the Sun's reaction to produce clean, near-limitless energy, pursued for 70 years but never commercialised at scale) is commercialised. Major genetic diseases are eliminated or prevented. Economies are radically decarbonised. Universal basic income exists in various forms in wealthy countries. Probability estimated by serious futurologists: 20 to 30%.
Median scenario: the difficult transition. The benefits of AI are massive but unevenly distributed. Part of the world's population (estimated at 2 to 3 billion) benefits from healthcare, increased longevity and material abundance. Another part remains excluded through lack of access to infrastructure or geopolitical conflict. Tension between blocs (USA, China, Europe, an expanded BRICS for Brazil, Russia, India, China, South Africa + new members) remains high. Several major crises (climate, economic, health) slow the trajectory. AGI exists and is in use, but remains under human control. Estimated probability: 40 to 50%.
Pessimistic scenario: drift. One or more of the major threats (loss of control of an AGI, AI-augmented conflict between powers, environmental collapse accelerated by AI's energy consumption, extreme concentration of power among a few companies) materialises. The resulting global crisis sets life expectancy back, fragments economies, and reverses several of the expected advances. Estimated probability: 15 to 25%.
You'll notice that the three scenarios combined come to 75–105%. That's deliberate. No serious futurist commits to a closed total. There is always a margin for "totally unforeseen events" (the "black swans", a term popularised by essayist Nassim Nicholas Taleb to describe rare, unpredictable events with high impact) that nobody can anticipate.
The philosophical questions nobody can dodge
Over a 30-year horizon, several questions become unavoidable and deserve to be thought about now.
Will we live forever? No, probably not. Current research is aiming at extending healthspan (the length of life spent in good health, as opposed to lifespan, which means total length of life), not immortality. Living 120 or 150 years in good health is a credible goal over 30 years. Living 500 years is science fiction. Living forever by uploading your consciousness remains largely speculative and probably impossible given what we currently know about biological consciousness. As we explored in our article on AI consciousness, science cannot even precisely define what consciousness is.
An end to disease? Most of the serious diseases known today will probably be treatable or preventable by 2056, but new diseases will emerge (bacterial resistance, environment-related conditions, rare genetic diseases that are unknown today). The phrase "end of disease" is a slogan, not a medical reality. Massively more effective medicine, yes.
Space exploration? Mars will probably be visited and probably inhabited by a permanent human presence by 2050. Beyond the solar system, it's less certain: physical distances make crewed interstellar travel extremely difficult. AI will help design intelligent autonomous probes capable of exploring exoplanets (planets outside the solar system), but humanity itself will probably remain confined to the solar system throughout the 21st century.
What place is there for humans in all this? This is the most important question, and the one whose answer depends on us. AI has no intrinsic agenda: it is humans who decide what to do with it. If societies reinvest in education, creativity, human connections and social support, humans keep their place. If they let AI mechanically replace humans without redistributing the benefits, societies fracture. The choice is political, not technological.
What to take away today
None of these scenarios is set in stone. But several things are already visible, and it's better to keep them in mind for the next 18 months.
First point: the acceleration is real and ongoing. As we wrote in our article on unmanaged acceleration, what seems impressive today will look rudimentary in 12 months. What seems impossible today will be commonplace in 5 years.
Second point: the opportunities go to those who adapt quickly. The startups that succeed in 2026 (Cursor, ElevenLabs, Anthropic) are those that understood quickly and executed quickly. The same logic will apply to individuals, to SMEs and to nations over the next 30 years.
Third point: the policy choices of 2026-2030 will strongly shape the future. AI regulation, redistribution of productivity gains, investment in research, training for populations, global governance: all decisions being taken now that will determine which scenario we tip into over 30 years.
Fourth point: foresight is not an excuse for inaction. Too many people tell themselves "anyway, AI is going to change everything, what's the point of making plans". The exact opposite is what's needed. Periods of rapid transformation reward those who act, not those who wait. The industrial fortunes of the 20th century were built in the years 1900-1920, not during the decades of stability.
To go further, you can read our article on the medicine of the future, which details concrete advances in cell regeneration, and our analysis of scientific discoveries made by AI, which explores what models can already invent in our fundamental knowledge. Nobody reads the future. But those who build it, at their own scale, starting today, are the ones who will have the fewest surprises tomorrow.