Artificial general intelligence has long been described as one of the technology industry’s most important future milestones. However, the closer AI systems appear to move towards broad human-level capabilities, the more difficult AGI has become to define.
AI leaders continue to predict significant advances over the next several years. Yet they do not necessarily agree on what AGI means, how it should be measured or which capabilities would prove that it has arrived.
Some executives now prefer terms such as “powerful AI” or “superintelligence.” Others describe intelligence as a spectrum rather than a single threshold that will be crossed on a particular day.
Recent reporting from Bloomberg highlights this growing uncertainty around one of the AI industry’s most widely discussed goals.
Artificial general intelligence generally refers to an AI system capable of performing a broad range of intellectual activities at or beyond human level.
Today’s AI systems can already produce content, write software, analyse information, use digital tools and assist with scientific research. However, their performance can be inconsistent. A model may succeed at an advanced technical problem while failing on a seemingly simple task involving context or common sense.
AGI is usually expected to demonstrate abilities such as:
The difficulty is that there is no universally accepted test covering all these capabilities.
The disagreement is not limited to researchers. Leaders of major AI organisations also describe the goal in different ways.
Open AI has historically described AGI in relation to systems that can outperform humans across most economically valuable work.
Sam Altman has also argued that the term has become imprecise as AI capabilities have developed. His public discussion has increasingly extended beyond AGI towards superintelligence—a stage at which AI would significantly exceed human capabilities.
Open AI has emphasised that the transition towards more powerful intelligence may occur gradually rather than through one clearly identifiable event. Its published approach also stresses the need for governance, safety evaluations and broad distribution of AI’s benefits. OpenAI’s AGI planning framework!
Anthropic CEO Dario Amodei frequently uses the term “powerful AI” instead of relying exclusively on AGI.
His description focuses on systems with broad intellectual capabilities comparable to highly skilled experts across fields such as biology, engineering, computer science and mathematics.
Amodei has suggested that systems resembling a large concentration of highly capable experts could emerge as early as 2026 or 2027, while acknowledging the substantial economic and security consequences that would follow. Anthropic’s statement on powerful AI.
Forecasts vary because experts make different assumptions about how AI will develop.
Important variables include:
Training advanced systems requires enormous computing resources, specialised chips, data-centre capacity and electricity.
Continued infrastructure growth could accelerate AI development, while hardware, energy or supply-chain limitations could slow it.
AI systems depend on high-quality training information and feedback.
Developers are exploring synthetic data, simulations and AI-generated training environments as easily accessible human-created data becomes less sufficient for frontier development.
Larger models have produced substantial improvements, but scaling existing methods may not be enough to achieve reliable general intelligence.
New architectures, memory systems, reasoning methods or learning techniques may be required.
AI agents can use software, complete multi-step tasks and interact with external systems.
Progress in agent reliability could be central to AGI because general intelligence would need to translate reasoning into sustained, useful action.
An AI model must do more than succeed occasionally.
Systems used in healthcare, finance, cybersecurity, engineering or public infrastructure need dependable performance, clear safeguards and predictable behaviour.
Digital intelligence does not automatically translate into physical ability.
Robots must process sensor information, navigate uncertain environments and interact safely with people and objects. This makes general intelligence in the physical world a separate and potentially slower challenge.
The AGI debate is not simply about terminology.
Definitions influence how companies, governments and the public prepare for advanced AI.
A clear and measurable description of AGI could affect:
Some commercial agreements may even connect legal or financial conditions to whether AGI has been achieved. An unclear definition can therefore create practical consequences beyond academic debate.
Businesses do not need to wait for a formal AGI announcement to prepare for more capable AI.
The immediate priority is understanding how current and emerging systems can change operations.
Organisations may increasingly use AI for:
Companies should evaluate where AI can create measurable value while establishing controls for accuracy, privacy, security and human accountability.
More powerful AI could automate a growing number of cognitive tasks. However, its employment impact is unlikely to be identical across every profession.
Routine and highly structured work may face greater exposure. Jobs involving responsibility, interpersonal trust, physical activity, domain expertise or complex judgment may change differently.
Demand could grow for professionals who can:
The shift may centre on the redesign of work rather than the immediate disappearance of entire occupations.
The uncertainty around AGI does not reduce the importance of preparing for advanced AI.
Professionals can focus on practical capabilities that are already becoming valuable:
Professionals who combine technical knowledge with expertise in healthcare, finance, insurance, manufacturing, retail or another industry may be especially valuable.
The debate around AGI is shifting from “When will it arrive?” to “What capabilities would prove that it has arrived?”
That is a useful change.
Predictions can encourage preparation, but dates alone do not provide businesses or professionals with an actionable strategy. Measurable capabilities, real-world reliability and responsible deployment matter more than a label.
The most important developments may occur gradually. AI agents could take on larger portions of software development, research, analysis and business operations long before the industry reaches agreement on whether those systems qualify as AGI.
For professionals, the practical response is to develop AI literacy, strengthen complementary skills and learn how to evaluate automated outputs. For organisations, it means building governance and security frameworks that can adapt as models become more capable.
AI leaders broadly agree that more powerful artificial intelligence is approaching, but they continue to disagree about the meaning and timeline of AGI.
Open AI has focused on economically valuable work and the progression towards superintelligence. Anthropic describes powerful AI as a concentration of expert-level intellectual capability. Google DeepMind examines AGI and superintelligence as stages along a broader continuum.
These perspectives point to the same conclusion: AGI may not arrive as one universally recognised moment.
Its effects could emerge through a sequence of advances in reasoning, autonomy, software use, scientific research and workplace productivity.
The question may therefore be less about declaring that AGI has arrived and more about whether society is prepared for increasingly general and powerful AI systems.